Aishwarya Srinivasan

AI@Fireworks AI | Global Speaker(100+ talks)| EB1-A Recipient| 570k+ Followers| AI Engineer| GenAI/LLM Inference| LinkedIn Top Voice| Author-What's Your Worth?| Startup AI Advisor| Angel Investor| Ex-Google,Microsoft,IBM

𝗙𝗼𝗿 𝗽𝗮𝗿𝘁𝗻𝗲𝗿𝘀𝗵𝗶𝗽 𝗶𝗻𝗾𝘂𝗶𝗿𝗶𝗲𝘀, 𝗰𝗼𝗻𝘁𝗮𝗰𝘁 𝗺𝘆 𝘁𝗲𝗮𝗺 𝗮𝘁 💌 partnerships@illuminate-ai.org

I write about:
- AI engineering
- Applied machine learning
- Generative AI Applications
- LLM/ LLMOPs (Training, post-training, inferencing)
- Fundamentals of Data science/ ML/ AI
- ML roles interview tips
- Learning roadmaps (beginner to advanced)
- Importance of personal branding for career growth

**All the posts reflect my own views and do not represent my employer.**

AI Advisor || Responsible AI Advocate || LinkedIn Top Voice - Data & AI  || Top 10 AI Influencer || EB1-A Receipt


Bio: 
Aishwarya currently leads the Developer Relations - Growth at Fireworks AI. She was previously at Microsoft for Startups as a Senior AI Advisor to help Y Combinator startups build machine learning solutions, leveraging core Microsoft/ OpenAI products. Prior to this, Aishwarya was working as a Data Scientist in Google Cloud and before that an AI & ML Innovation Leader at IBM Data & AI, where she was working cross-functionally with the product team, data science team and sales to research AI use-cases for clients by conducting discovery workshops and building assets to showcase the business value of the technology. She holds a post-graduate in Data Science from Columbia University. 

Aishwarya has been awarded Trailblazer of the Year by Women in AI in 2022 and Women of Influence by Business Journal in 2022. She was awarded EB1A - extraordinary ability visa for her contributions in the AI field.

She is an advocate for open-source technologies, and was an open source Developer Advocate for Deepchecks and Lightning AI, and a contributor to Scikit Learn. 


She holds a Patent Award (2018) for developing a Reinforcement Learning model for Machine Trading.

She has over 500,000+ follower base on LinkedIn and actively organizes events and conferences to inspire budding data scientists. She has been spotlighted as a LinkedIn Top Voice 2020 for Data Science and AI, which features Top 10 Machine Learning influencers across the world. 

She is an ardent reader and has contributed to the scholastic community. To spread her knowledge in the space of data science, and to inspire budding Data Scientists, she actively writes blogs related to machine learning on LinkedIn and Instagram: www.instagram.com/the.datascience.gal

Besides being a data junkie, she is a fitness fanatic who is into martial arts (Krav Maga) and yoga.

Senior AI Advisor

If you're feeling overwhelmed with how fast AI is evolving, you're not alone.

Every day there’s a new paper, a new framework, a new agent loop, and it’s easy to feel like you’re falling behind.

But the good news is that you don’t need to learn everything all at once. What you need is structure.

So I put together a 10-level AI Agents Learning Roadmap that takes you from foundations to production, layering your learning in a way that’s actually doable.

💡My recommendation: spend 2–3 weeks on each level. Learn the concepts, implement small projects, and build your intuition.

If you're moving faster or slower based on time or experience, that’s okay too.
And when something new drops? That can be your Level 11. Don’t let “newness” derail your plan. Just start here.

👇 Here’s the roadmap:
🔖 Level 1: GenAI & Transformer Foundations
 Tokens, embeddings, transformers, decoding, and inference with open-weight models.

🔖 Level 2: Prompting & Language Model Behavior
 Prompt types (CoT, ReAct, ToT), decoding strategies, context design, and adversarial prompting.

🔖 Level 3: Retrieval-Augmented Generation (RAG)
 Chunking, embeddings, vector DBs, RAG pipelines, and RAG evaluation.

🔖 Level 4: LLMOps & Tools
 LangChain, LangGraph, Dust, CrewAI, tool use, function calling, and synthetic data.

🔖 Level 5: Agents & Agent Frameworks
 Agent types, memory, planning, LangChain agents, LangGraph loops, and evaluation.

🔖 Level 6: Memory, State & Orchestration
 Vector and symbolic memory, episodic vs persistent state, memory compression.

🔖 Level 7: Multi-Agent Systems
 Hub-and-spoke vs decentralized, message passing, collaborative agents, agent teams.

🔖 Level 8: Evaluation & Reinforcement Learning
 LLM-as-a-Judge, RLHF, RLVR, reward modeling, and self-correcting loops.

🔖 Level 9: Protocols & Safety
 MCP, A2A, safety alignment, guardrails, traceability, and autonomous policy updates.

🔖 Level 10: Build & Deploy
 FastAPI, Streamlit, GGUF, QLoRA, caching, monitoring with LangSmith, Arize, Trulens.

📌 Bookmark this.
🛠️ Build something after every level.

And if you're wondering what tools to explore along the way → Start with Hugging Face (to explore LLMs and SLMs), you can use Ollama (to run SLMs on your laptop, like Phi-4, TinyLlama), or Fireworks AI (to run LLMs via endpoint, like Qwen 3, Kimi K2, DeepSeek R1), then explore LangChain & LangGraph (these two tools will  teach you a lot), then you can move into learning Agentic frameworks like CrewAI, AutoGen. 

💻 Pro-tip: Start with cookbooks!

〰️〰️〰️
Follow me (Aishwarya Srinivasan) for more AI insight and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

Hundreds of AI minds gathered in Central Park on a chilly Saturday morning, it was incredible :)

The past weekend in New York felt like one of those moments where community completely surprises you in the best way.

I was traveling to the city for a work event, and my friend Pranjal and I had a thought- What if we host a small AI meetup on Saturday morning in Central Park? Nothing fancy, just a space for people to connect.

So we put up a quick Luma link and shared it on LinkedIn. We expected maybe a handful of people. Instead, the sign ups crossed 450+. 

We put a small table with coffee, muffins, cookies and some cool swag- thanks to Teachable. We weren’t sure how it would go. We just hoped people would show up.

What happened next honestly moved me.

We reached the park at 9.20am to set up for a 10am meetup, and people were already there. In the cold. On a Saturday morning. People had traveled from Boston, Maryland, upstate New York, New Jersey, even Philadelphia. I kept talking to folks thinking, you really came all this way… for this?

The meetup was meant to last two hours. People stayed until 3pm! Almost six hours of conversations about AI, careers, ideas, and everything in between. No microphones. No agenda. Just humans connecting because they care about the same things.

This experience reminded me of something I tend to forget- community doesn’t need perfect planning. It needs intention. It needs genuine people. And it needs a space where everyone feels welcome.

I’m grateful to every single person who showed up. You made this morning unforgettable, and you reminded me why I love building in this space.

Shout-out to Aditya Suresh and Pranam Daga for helping us with putting this together ❤️

There is a quiet moment every November when life slows down just enough for reflection. I never grew up celebrating Thanksgiving. Back home in India, I didn’t even know what it meant. But after moving to the US, watching my friends celebrate, watching sitcoms where families gathered around food and gratitude, I started understanding the deeper meaning behind it.

A moment to pause.
A moment to look around.
A moment to look back.

This year, gratitude feels a little different for me. A little heavier in the best possible way.

I feel grateful for a life where I get to wake up every day and do work I genuinely love. AI is chaotic and intense and fast. Yet it gives me energy. It gives me purpose. And I feel lucky to be part of this ecosystem during a time when everything is shifting.

I feel grateful for the people who hold me steady.
A mom who pushed me when the world expected her to play it safe. A mom who defied every norm to carve space for me to choose ambition without apology.

A husband who is involved in my life in the most beautiful way supportive is honestly an understatement. He is a true partner in every single thing I do. He listens to every idea, every frustration, every new plan I dream up and he backs me with a kind of belief that strengthens my own. Chinnu amma and Chinnu appa who patiently listen to my ten year entrepreneurial plans and cheer for me with the kind of warmth only family can give.

Mentors and colleagues who let me thrive creatively who challenge me in the right ways who make every room I walk into feel lighter and more possible.

I feel grateful for health. For a body I get to take care of. For the time I carve out to move, to eat well, to nurture myself. It took years to realize that taking care of your body is also taking care of your dreams.

And I feel grateful for the little girl I used to be the one who sat with a notebook and wrote her goals on every page the one who kept believing she would get there someday. If I could talk to her today, I would tell her that she is doing great and that her stubborn belief in herself was worth it.

Thanksgiving may not be something I grew up with, but it has become a reminder to pause and appreciate the rarest gift of all the good people around you. People who are genuinely supportive, genuinely positive, and genuinely rooting for you.

I am grateful for all of it.
And grateful for everyone who has been part of my journey so far.

If you are reading this, I hope you take a moment to celebrate the smallest wins, the unexpected joys, and the people who make your life feel expansive.

li_at

JSESSIONID

What an electrifying afternoon! Absolutely delighted to have joined the Mayfield AI List honorary lunch at the beautiful Golden Gate Club in San Francisco.

The room was filled with the brightest minds in the AI VC world. it was a privilege to meet and listen to inspiring leaders like Navin Chaddha, Aparna Chennapragada and Garry Tan. 

I had the opportunity to host a focused roundtable, diving deep into the current state of the industry. We tackled the crucial discussion around Perception vs. Reality of AI, analyzing current hype cycles, and zeroing in on investment opportunities that hold real long-term potential.

It’s moments and small-group settings like these, where leaders can candidly discuss where the real value lies beyond the buzz, that drive the industry forward.

Huge thanks to the Mayfield and Startup Grind team (special shoutout to James Gee and Fadia Frangie Doumat) for hosting and for the award ♥️

No alternative text description for this image

Wrapped up AI Dev 25 hosted by DeepLearning.AI

More than 1400 AI developers gathered in one space, and the energy was something you could feel instantly. Meeting so many builders and hearing what you’re creating reminded me why this community moves the needle the way it does.

Hearing Andrew Ng talk about the future of work was a standout moment. The way he encourages everyone on his team to code, regardless of background, is exactly the kind of mindset shift this era needs. AI literacy is becoming a baseline skill, and he articulated that perfectly.

Our panel with Michael Kearns, Hatice Ozen, Kay Zhu, and Josh Gordon covered how the ecosystem is evolving across the stack, from Groq , Google, and Amazon Web Services (AWS)  to Fireworks AI and how teams like Genspark are using RFT on Fireworks AI in production workflows. We also touched on multi-agent systems, practical best practices, and where developer tooling still needs to mature.

I also loved my conversation with Michael Kearns on responsible AI and how that thinking has evolved from classic ML days to the generative era. His perspective at the intersection of academia and industry always leaves me thinking differently.

Grateful for the conversations and connections today. If we met, thank you for stopping by. If we didn’t, I’m always happy to keep the discussion going. There’s a lot to build, and we’re still early.

Thank you DeepLearning.AI team for having me ❤️

When you’re an AI engineer, the worst bottleneck isn’t writing code. It’s navigating everything around the code.

Last week I tried out Rovo Dev from Atlassian and it really changed how I approach dev work, plus how I see AI agents changing the landscape of software engineering.

𝗪𝗵𝗮𝘁 𝗶𝘀 𝗥𝗼𝘃𝗼 𝗗𝗲𝘃?
RovoDev is an AI agent built to live inside your dev workflow, including your code, your issues (let’s say in Jira), your docs (let’s say in Confluence), your pull requests. 

It helps you move from idea → code → deployment with fewer context switches.

𝗛𝗲𝗿𝗲’𝘀 𝘄𝗵𝗮𝘁 𝗜 𝗻𝗼𝘁𝗶𝗰𝗲𝗱 𝗮𝗳𝘁𝗲𝗿 𝘂𝘀𝗶𝗻𝗴 𝗶𝘁:
→ I asked in plain English: “What does this module do? Where are its dependencies?” and got a clear breakdown, no digging through 3 dozen files.
→ I linked my GitHub/Bitbucket + Jira and was able to surface relevant issues, test failures and comments in one context instead of switching tools.
→ The CLI experience (yes, in my terminal) allowed me to trigger agents to draft tests, summarise PRs, or suggest refactoring ideas, without leaving my code. 

𝗪𝗵𝘆 𝗶𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀?
What I realized after using Rovo Dev is that the real time-sink in engineering isn’t writing code, it’s in fact more about re-orienting yourself.

Jumping between Jira, Confluence, and your repo just to understand context.
Trying to figure out why something was built the way it was.
And then losing your train of thought before you can even start solving the problem.

Well, that's where Rovo Dev actually helps close that gap.

If you’ve ever joined a new project and spent hours just piecing together what’s going on, or if you constantly find yourself switching between tools to get one small thing done, you should try this out: fandf.co/42MTjr8 

Thank you Atlassian for partnering with me on this post, and giving me access to Rovo Dev to test it out :)

#AtlassianPartner

This is one of the best resources for aspiring AI engineers, that covers all the basics - advanced in classroom teaching: https://lnkd.in/dgr5zEve

Highly recommend saving this so you can come back to it later.

Goldmine for aspiring AI Engineers

Lately, I’ve noticed something interesting in the AI startup world.

The ones growing the fastest aren’t building the next big model.

They’re building agentic systems.
AI that doesn’t just talk, it does.
Copilots that summarize meetings.

Agents that predict churn or handle internal ops automatically.

These aren’t demos anymore. They’re real products driving revenue.

What separates the ones that scale from the ones that stall comes down to one thing : Data.
Every great agent lives and dies by how it manages memory, retrieval, and reasoning speed.
And that’s exactly why so many founders I work with build on MongoDB Atlas.

It’s not just a database.
It’s the data layer that brings structured, unstructured, and vector data together so your AI can actually reason in real time.

If you’re a startup building AI that acts, not just talks, go check out the MongoDB Startup Program: https://fandf.co/493VsD1

#MongoDBPartner

🚨Context Engineering for GenAI Workshop happening tomorrow!

In this session, we’ll dive deep into:
→ What context really means in LLMs and how it shapes performance

→ The toolkit that helps you engineer it better: Summarization, Pruning, Query Rewrite, RAG, Context Offloading, Isolation, and Memory

→ How to optimize for accuracy, cost, and reliability across use cases

→ A live demo showing how these ideas come together in real time

This workshop is for every engineer who wants to move beyond prompt tuning and start thinking like a system designer.

🧩 Context Engineering
📅 November 8
🕐 1.5 hours
🎙️ with Arvind Narayanamurthy & me

🎟️ Save your spot: https://lnkd.in/dvFmMiRr

If you’re an AI engineer today, you should absolutely get familiar with Model Context Protocol (MCP) and Responses API. These two are becoming the backbone of modern agentic systems.

Let’s unpack why 👇

1️⃣ MCP: The Connector Layer

Model Context Protocol (MCP) standardizes how models connect to external tools, APIs, and live data.

Think of MCP as the wiring system that allows your LLM to talk to the outside world, like tools, databases, vector stores, or your company APIs.

When you build or host your own MCP server, you’re giving your model new abilities to reason, act, and fetch context beyond what’s in its prompt.
This is what turns a plain LLM into an agent.

If you’re working on production-grade AI systems, you’ll often need to:
→ Integrate custom tools or data sources
→ Maintain context across multiple steps
→ Ensure reliable tool invocation and streaming

MCP makes all that possible.

2️⃣ Fireworks AI Responses API

Once your MCP setup is ready, the Responses API is how you connect it to the model.

It’s a next-generation SDK that powers conversational AI. It supports:
→ Stateful conversations (previous_response_id)
→ Real-time streaming for interactive apps
→ Tool calling and safe execution

The Fireworks Responses API follows the same schema and format as OpenAI’s Responses endpoint.

So if you’ve been using OpenAI models, you can switch to open-source models on Fireworks AI with almost zero code change.

You’ll unlock faster inference, lower latency, and broader model choice, including open models like Qwen, Mistral, Llama, and Gemma hosted natively on Fireworks AI.

You can even mix and match:
✅ Use MCP to extend model capabilities
✅ Use Responses API to orchestrate reasoning, tool use, and streaming

That’s the magic combo powering many of today’s most advanced agentic systems.

3️⃣ Why This Matters

Agentic systems aren’t built with just a model call.
They’re built with context, memory, and actions. MCP and the Responses API make that orchestration reliable, stateful, and production-ready.

If you’re serious about building enterprise-grade AI apps, start experimenting with:
→ MCP Examples (connecting to MCP servers and invoking tools)
→ Conversation Chaining (using previous_response_id to continue dialogue)
→ Streaming Output (for responsive real-time interfaces)
→ Ephemeral Mode (for privacy-friendly, non-persistent interactions)

Happy Building 🚀
〰️〰️〰️
Follow me (Aishwarya Srinivasan) for more AI insight and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

If you’re an AI engineer or an AI leader looking to level up your skills in LangChain and LangGraph, this one is for you.

I’m teaming up with Arvind Narayanamurthy  for a beginner–intermediate, 1.5 hour hands-on workshop where you’ll learn how to move from simple prompts to real AI applications and agentic workflows.

Why this matters
- Most engineers know how to prompt. Few know how to orchestrate.
- LangChain and LangGraph are becoming the backbone of production AI systems.
- If you want to build or lead AI engineering teams, you need these mental models.

What you’ll learn
- Foundations of LLM orchestration and where LangChain fits
- Essential building blocks: chat models, prompting patterns, structured outputs, LCEL
- Tool calling with MCP adapters
- RAG basics end to end
- LangGraph workflows and agentic architectures
- Two live labs: a micro-app and a simple agent
- Quick dive into LangSmith for debugging

The workshop is recorded, and everyone gets the replay.

If you want practical, real-world skills to build modern AI systems, this is the one to sign up for.

Grab your seats: https://lnkd.in/eW2WZdcf

Are you wondering where the AI industry is actually headed? 
Every week brings a new framework, a new toolkit, a new acronym. But beneath all the noise, one truth is becoming clearer: the real moat isn’t the model, it’s the combination of data and AI. 
 
AI is the toolkit. 
Data is the substance. 
AI gives us reasoning power. 
Data gives us understanding. 

That connection came to life at Snowflake BUILD 2025, where the  focus wasn’t on hype, it was on grounding AI in reality. 
 
We saw how Snowflake Intelligence lets teams ask complex questions in natural language and get trusted answers from governed data. How the Cortex Agents API enables developers to build domain-specific agents that reason and collaborate. And how the Managed MCP Server securely bridges external AI systems to enterprise data, turning “agentic AI” from an idea into something production-ready. 
 
Even unstructured data, the 80 % of information hidden in PDFs, chats, and documents, is now first-class and accessible through simple AISQL queries. 
 
This feels like a turning point. 
The next era of AI won’t be defined by bigger models, but by deeper integration between data and reasoning. 
 
AI without data is just potential. 
Data without AI is just storage. 
Together, they become intelligence. 
 
#SnowflakePartner #SnowflakeBUILD

If you’re an AI engineer trying to understand agentic systems, keep reading 👇

A single-agent system relies on one model to do everything: reasoning, planning, execution, and validation. It’s efficient for simple, sequential tasks but quickly breaks down as workflows grow more complex or interdependent.

A multi-agent system, on the other hand, distributes intelligence across specialized agents that collaborate. Each has its own role, memory, & context - and together, they solve problems that a single model can’t handle alone.

But here’s what most people miss: there’s no standard architecture.

A multi-agent system is like Lego blocks, you design it for your use case.

You can build:
→ a supervisor agent managing sub-agents
→ a hierarchical setup with layered control
→ or a hybrid model combining centralized planning and peer coordination
The challenge isn’t in building agents, it's in managing how they talk to each other.

Three reasons multi-agent systems often fail:

❌Memory fragments across agents.
Each agent keeps its own context. When one agent hands work to another, subtle details get lost or duplicated. Without a shared memory layer or context synchronization, agents start “forgetting” what the others already know.

❌Operational costs multiply.
What costs $0.10 for a single agent might cost $1 when ten agents coordinate. Every message, validation step, or context handoff adds latency, token cost, and complexity. At scale, coordination overhead often outweighs the benefit of specialization.

❌ Write conflicts cascade.
If multiple agents write to the same source of truth: a shared database, knowledge base, or document, their changes can collide. One agent updates a schema, another updates logic on top of the old version, and suddenly the system’s state fragments.

TL;DR: Not every workflow needs a multi-agent setup.

When tasks are independent, read-heavy, and parallelizable, multi-agent systems shine. But for sequential logic and tight dependencies, a single-agent pipeline is still faster, cheaper, and easier to maintain.

As always, Galileo has published a new phenomenal free handbook- Mastering Multi-Agent Systems.
 
If you want to go deeper into how these systems work, why they fail, how to evaluate them, explore different architectures, and even build LLMOps around them, this is one of the best reads out there: https://lnkd.in/diptdEC2

#GalileoPartner

Multi-Agent AI System Cheatsheet 

I learned Design Thinking on my first day as a data scientist at IBM, and it quietly became one of the most useful mindsets in my AI career. It taught me to slow down, understand the real problem, and build solutions that actually land with users.

Here’s how I still use it today:

→ Solve the right problem
Most AI teams jump into modeling too early. Design Thinking forces you to reframe, question assumptions, and check if the problem is even defined correctly.  

→ Diverge before you converge
Explore many possibilities, then narrow down. This prevents tunnel vision and leads to better architectures, better experiments, and better product decisions

→ Use the 2x2 grid
Am I exploring or deciding?
Am I thinking about the problem or the solution?
This simple check saves me from going down the wrong path.  

→ Observe behavior, not opinions
The real insights come from watching users, not asking them. The same applies when evaluating agents, prototypes, or internal tools.  

→ Prototype fast, assign ownership
Ideas die when nobody owns them. Execution is where Design Thinking turns into impact.  

This mindset shaped how I build AI systems then, and it still shapes how I build them now.

♻️Share this with your network

Design Thinking Framework

Last week I came home to a little surprise from OpenAI filled with their new books, 100 Chats for Studying, Career, and College Life. 

I flipped through them right away and loved how visual and genuinely interesting they’ve made each section. It feels approachable, practical, and fun in a way most tech books don’t even try for.

If you want to explore the prompts yourself, you can check them out at this link: https://lnkd.in/ddycaDAg

It’s a pretty cool resource to keep on your desk as you build, study, or just experiment with new ideas.

Here is my absolute fav one which I wish I could have used as a student 👇  
I want to understand [concept]. Explain it with an analogy to real-world items.

Let me know in the comments below which ones is your favorite?

If you’re getting started as an AI engineer, this one’s for you 👇

Here are 12 things you need to learn if you want to become a real AI engineer (or level up as one):

1️⃣ LoRA & qLoRA fine-tuning
Learn low-rank adaptation to fine-tune massive models while keeping GPU and memory usage in check.
https://lnkd.in/dxAfNJjP
https://lnkd.in/dcfhk4YB

2️⃣ Reinforcement Learning from Human Feedback (RLHF)
Understand how to align model behavior with human expectations using curated reward signals.
https://lnkd.in/da4qf5JV
https://lnkd.in/drEbQ4x2

3️⃣ Reinforcement Learning from Verifiable Rewards (RLVR)
Use simulated or programmatic reward models when human feedback is too expensive.
https://lnkd.in/dFBgDAyd

4️⃣ RAG Fundamentals
Learn how to build Retrieval-Augmented Generation systems- ingest documents, create embeddings, stitch context, and prompt LLMs.
https://lnkd.in/dFkgD6RQ
https://lnkd.in/d7v6DzEG

5️⃣ Advanced RAG Architectures
Master multi-hop retrieval, hybrid sparse-dense search, and complex reranking for multi-document use cases.
https://lnkd.in/dUrAz2D3

6️⃣ Embedding Models & Vector Databases
Benchmark sentence encoders, tune indexes (FAISS/Pinecone), and understand latency vs recall trade-offs.
https://lnkd.in/d8zeWbmn
https://learn.pinecone.io/

7️⃣ Prompt Engineering & Chain-of-Thought
Design prompts that guide reasoning step-by-step, manage templates, and dynamically inject context.
https://lnkd.in/djbJD578
https://lnkd.in/d9M75H_i

8️⃣ MCP & Agent-to-Agent Integration
Use Model Context Protocol (MCP) to coordinate LLMs with APIs and other agents for seamless A2A handoffs.
https://lnkd.in/dfjvxhvP
https://lnkd.in/dJ5NWuaK

9️⃣ Long-Context Window Strategies
Handle 100k+ token sequences using smart chunking, hierarchical retrieval, and memory-augmented architectures.
https://lnkd.in/djgkXGaZ

🔟 Model Efficiency: Quantization & Distillation
Shrink models with post-training quantization (4/8-bit), pruning, and knowledge distillation for faster, cheaper inference.
https://lnkd.in/d2iurEyv

1️⃣1️⃣ LLMOps & Model Monitoring
Build CI/CD pipelines for LLM deployment, track drift, monitor hallucinations, and set up alerting for cost or performance issues.
https://lnkd.in/dxik45e6

1️⃣2️⃣ Evaluation & Benchmarking
Measure with perplexity, ROUGE, F1, and human evaluations. Automate benchmarks to compare models at scale.
https://lnkd.in/dP_NJXg9

〰️〰️〰️〰️
♻️ Share this with your network 
🔔 Follow me (Aishwarya Srinivasan) for data & AI insights, and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

If you’re getting started with building AI agents, this Agentic AI Toolkit Cheatsheet will come in handy 👇

You’ll quickly realize it’s not just about choosing a model. You need orchestration, memory, retrieval, evaluation, and a solid deployment stack.

And if you’re not careful, your architecture turns into a mess of disconnected tools.

That’s why I put together this Agentic AI Toolkit Cheatsheet.

It organizes the full stack of what goes into building production-grade AI agents- from model hosting to feedback loops, so you can stop guessing what’s missing in your pipeline and start shipping actual systems.

Here’s what it covers:

🔧 Foundation Models & Hosting
🎛️ Model Customization (SFT, RFT, RLHF)
🤖 Agent Frameworks
🧩 Orchestration & Workflow
📥 Data Ingestion & Preprocessing
🧠 Embedding & Retrieval
🧷 Memory & Knowledge Storage
🔌 Tool Integration
🧪 Evaluation & Monitoring
🖥️ User Interfaces
🚀 Deployment & Hosting


💡 A few tips if you’re working on your first agentic stack:

→ Choose one tool per layer to start, depth over breadth.
→ Build iteratively and introduce evaluation as early as possible.
→ Use event-driven workflows to manage complexity.
→ Think about memory and feedback as first-class components, not afterthoughts.


📌 Bookmark this.
📚 Use it to audit your stack.
🛠️ Build with it.
📤 Share it with a teammate who’s lost in the sea of GenAI tooling.

〰️〰️〰️
Follow me (Aishwarya Srinivasan) for more AI insight and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

This Week of AI is here⚡️
Another packed week in the world of AI- here’s your quick TL;DR of everything that happened across the AI ecosystem this week 👇

- Apple Forges Billion-Dollar AI Alliance with Google for Siri Overhaul
- Anthropic Foils First Documented Large-Scale Autonomous AI Cyberattack
- World Labs Unveils Marble, A Multimodal World Model for Spatial Intelligence
- OpenAI Rolls Out GPT-5.1 with Adaptive Reasoning and Dual Personalities
- Microsoft Creates MAI Superintelligence Team Led by Mustafa Suleyman
- Google Launches ‘Agentic Checkout’ and AI Shopping Tools for Holiday Season
- IBM Joins OpenSearch Software Foundation as Premier Member

♻️ Share this with your network
🔔 Follow me (Aishwarya Srinivasan) for no BS data and AI updates and insights

This week, the AI landscape was defined by blockbuster partnerships, a major infrastructure upgrade, and a chilling preview of autonomous cyber warfar

Open article: This Week in AI: Week of 16th November 2025 by Aishwarya Srinivasan • 3 min read

This Week in AI: Week of 16th November 2025

Thanksgiving week, yet a lot happening in this week in AI!
Catch everything you need to know here 👇

Vision-Language Models connect what AI sees with what it reads and reasons.

They’re the foundation of AI systems that can interpret charts, medical images, retail shelves, or product catalogs.

But a generic VLM doesn’t understand your domain’s visual language.
That’s where fine-tuning becomes essential.

𝐖𝐡𝐲 𝐟𝐢𝐧𝐞-𝐭𝐮𝐧𝐢𝐧𝐠 𝐚 𝐕𝐋𝐌 𝐦𝐚𝐭𝐭𝐞𝐫𝐬

A pretrained VLM already knows the basics of visual-text reasoning.
Fine-tuning helps it specialize for your domain.

→ In healthcare, it learns to detect anomalies in MRIs and X-rays.
→ In retail, it interprets shelf images and product layouts.
→ In enterprise, it extracts structured data from invoices and reports.

You’re not rebuilding intelligence, you’re refining perception to fit your use case.

𝐇𝐨𝐰 𝐋𝐨𝐑𝐀 𝐦𝐚𝐤𝐞𝐬 𝐢𝐭 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭

Full model fine-tuning is expensive and compute-heavy.
Low-Rank Adaptation (LoRA) keeps the base model frozen and trains only small adapter layers.

That means:
→ Faster training cycles
→ Smaller memory footprint
→ Lower compute costs
→ Domain adapters that are easy to swap in and out

You can maintain one base model and multiple lightweight adapters for each use case such as invoices, medical forms, or retail analytics.

𝐈𝐧𝐬𝐢𝐝𝐞 𝐚 𝐕𝐢𝐬𝐢𝐨𝐧-𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥

A VLM has three main components:
→ 𝐕𝐢𝐬𝐢𝐨𝐧 𝐄𝐧𝐜𝐨𝐝𝐞𝐫 converts pixels into visual tokens.
→ 𝐅𝐮𝐬𝐢𝐨𝐧 𝐋𝐚𝐲𝐞𝐫 combines visual and text context for reasoning.
→ 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐃𝐞𝐜𝐨𝐝𝐞𝐫 generates captions, summaries, or structured responses.

Each layer introduces potential failure modes like poor resolution, misaligned regions, or verbose hallucinations. Fine-tuning improves alignment and reliability across these components.

𝐓𝐡𝐞 𝐟𝐢𝐧𝐞-𝐭𝐮𝐧𝐢𝐧𝐠 𝐥𝐢𝐟𝐞𝐜𝐲𝐜𝐥𝐞

→ 𝐃𝐚𝐭𝐚 𝐝𝐞𝐬𝐢𝐠𝐧: Collect diverse, high-quality, clearly labeled visuals.
→ 𝐓𝐚𝐬𝐤 𝐝𝐞𝐟𝐢𝐧𝐢𝐭𝐢𝐨𝐧 : Choose the right setup: captioning, VQA, extraction, or localization.
→ 𝐋𝐨𝐑𝐀 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠: Train adapters for each domain efficiently.
→ 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧: Use both quantitative metrics and human review for grounding and accuracy.

Always evaluate across different slices such as document type, lighting, and template to surface hidden biases.

𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 𝐚𝐧𝐝 𝐫𝐞𝐥𝐢𝐚𝐛𝐢𝐥𝐢𝐭𝐲

Each domain adapter should have its own dataset lineage, version, and evaluation score.
Reliability requires attention to fairness, privacy, consistency, and uncertainty handling.
Fine-tuning doesn’t just improve accuracy, it strengthens governance and ethical alignment.

LoRA fine-tuning makes VLMs faster to adapt, cheaper to deploy, and more aligned with your real-world data. 

〰️〰️〰️
Follow me (Aishwarya Srinivasan) for more AI insight and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

Greg Coquillo

Product Leader @AWS | Startup Investor | 2X Linkedin Top Voice for AI, Data Science, Tech, and Innovation | Quantum Computing & Web 3.0 | I build software that scales AI/ML Network infrastructure

Product leader with a passion for solving business use cases such as Growth, Go-to-Market, and Price Optimization. I owned P&L for commodity-based products, and leveraged AI to increase the average customer contract values across business industries. As a Product Leader at Amazon, I own a roadmap that drives the adoption of AI products through the Media Automation charter, which supports Private Brands’ YOY growth. I work with Data Scientists, Software Engineers and Business Intelligence Engineers to leverage Machine Learning and Big Data Analytics to design predictive models that enhance business decision-making and performance to accelerate growth, reduce Time to Market of products, monitor cross-geo compliance and import requirements, and ensure customer safety.

I am always curious and eager to learn how new technology affects our daily lives, and I love to support communities on their quest to adopt Artificial Intelligence across industries. Feel free to contact me if you have Data Science projects that you’d like to be showcased on this platform for visibility.

🌐I am a Global Business Leader with 12 years of experience and skills including Product Management, P&L management, Global Networking, EBITDA Optimization, Pricing Management, Margin Management, Product Management, Program Management, Operations and Supply Chain Management, Lean Manufacturing, Six Sigma, and Continuous Improvement.

🧘🏽‍♂️Soft Skills include: Communication, Collaboration, Lateral Thinking, Adaptability, Inspiration, Innovation, Empathy, Listening, Diplomacy, Organization, Responsibility, Leadership...

🎬🎥🎞 I really enjoy watching movies, and if I could redo it all over, I’d be a Sci-Fi script writer so I can be bounded by the limits of imagination.

(Opinions are my own)

Product Management, Media Automation

Owner of the AI vision and roadmap for Private Brand’s Compliance Division. Driving adoption of new technology, allowing my customers to scale their processes and increase productivity. Core responsibilities include:

-Building cross-functional teams consisting of Data Scientists, Software Engineers, Data Engineers, Product Managers and Program Managers, to plan and implement sophisticated AI projects
-Manage the New Initiative Intake Process
-Gather Business requirement artifacts
-Communicate roadmap definition and prioritization
-Coordinate User Acceptance Testing
-Manage Launch Announcements
-Prepare and communicate annual planning artifacts

Building Agentic AI systems beyond connecting APIs or LLMs is complicated, but not impossible.

This architecture lays the foundation for how AI agents think, communicate, and improve, covering everything from testing and observability to deployment and memory management.

Here’s a breakdown of the key layers and components that make up a scalable Agentic AI Architecture :

1.🔸Decomposition
Break down complex systems by domain (e.g., Coding Agent, Data Agent), by cognitive capability (Reasoning, Planning, Execution), or by agent role (Planner, Executor, Memory Manager, Communicator).

2.🔸Communication
Enable message passing between agents using inter-agent protocols or A2A (Agent-to-Agent) orchestration.
Support both single-agent and multi-agent setups for small or distributed workflows.

3.🔸Deployment
Deploy agents in containerized or serverless environments using Docker or Modal.
Support orchestrators like CrewAI or AutoGen for collective intelligence in multi-agent workflows.

4.🔸Data & Discovery
Integrate knowledge bases (like vector databases for RAG), memory stores (FAISS, Redis, Pinecone), and APIs for dynamic data access.
Context is passed using Model Context Protocol (MCP) for structured and real-time reasoning.

5.🔸Testing & Observability
Validate workflows end-to-end, test reasoning logic, and evaluate performance under real conditions.
Monitor using Weights & Biases, LangFuse, and track metrics like latency and task success rate.

6.🔸UI & Style
Provide intuitive feedback loops through visualization layers, dashboards, and self-reflective modes.
Enable collaborative, proactive, and goal-driven reasoning among multiple agents.

7.🔸Security
Protect access with token-based authorization and data encryption.
Include Trust Layers for human-in-the-loop validation and Policy Enforcement for safe execution.

8.🔸Cross-Cutting Concerns
Handle configuration, secrets, and environment management.
Support flexible frameworks like LangChain, AutoGen, or CrewAI for runtime execution and modular design.

Agentic AI is the future of automation - where AI doesn’t just assist but collaborates and learns.

Save this post to understand the architecture that powers the next generation of AI systems

#AgenticAI

AI agents are great at working alone. However, they can also be effective when operating within networks of collaboration.

The real power of AI lies in how these agents communicate, coordinate, and complement each other’s abilities.

Orchestration patterns define this collaboration which turns isolated agents into intelligent, goal-driven systems capable of solving complex tasks together.

Here are 6 orchestration patterns that power the next generation of AI systems :

1.🔸Sequential Orchestration
Agents work one after another - each passing its result to the next.
Perfect for structured, step-by-step tasks like data pipelines or analysis chains.

2.🔸Concurrent Orchestration
Multiple agents work in parallel, handling different parts of a problem simultaneously.
This boosts speed, scalability, and overall task efficiency.

3.🔸Group Chat Orchestration
Agents communicate in a shared conversation, refining ideas together.
Useful for brainstorming, collaborative decision-making, or generating consensus.

4.🔸Maker–Checker Loops
One agent creates or proposes; another verifies, corrects, or approves.
Ideal for maintaining accuracy, compliance, and quality assurance in workflows.

5.🔸Handoff Orchestration
Agents transfer control dynamically based on context or expertise.
Best suited for adaptive workflows like support systems or multi-step automation.

6.🔸Magnet Orchestration
A central “manager” agent dynamically assigns tasks to the right specialists.
Enables fluid, context-aware collaboration across multiple intelligent agents.

AI Agents are evolving fast and orchestration is what makes them powerful.

Learning how these patterns can help you build smarter, more collaborative AI systems can be a huge advantage.

#AIAgent

AI Agent Orchestration Patterns

📚LLMs in the Enterprise are finally getting the playbook they deserve… LLMs in Enterprise by Ahmed Menshawy and Mahmoud Fahmy provides practical guides for building real AI systems that operate at scale.

Most teams talk about LLMs in theory. This book focuses on execution. It bridges foundational concepts with the hands-on design patterns that matter when you are integrating models into production environments.

Here are the insights that stood out👇

1.🔸Enterprise LLM integration is a data architecture problem
The book breaks down how to design pipelines, tune retrieval, and structure data so models operate with consistency and low latency in real workloads.

2.🔸Scaling LLMs requires pattern-level thinking
They go deep on architectural patterns that reduce complexity, improve efficiency, and streamline deployment. This includes RAG frameworks, fine-tuning strategies, segmentation techniques, and evaluation patterns that teams often overlook.

3.🔸Performance is not just about bigger models
The authors show how to optimize model behavior with advanced inferencing engines, contextual model customization, and monitoring systems that keep applications predictable.

4.🔸Real enterprise value comes from operational rigor
Security, fairness, transparency, and accountability are not afterthoughts. They are part of the design process, especially when LLMs touch business workflows and customer data.

5.🔸AI teams win by mastering both concepts and impact
The flow of the book reflects the real enterprise lifecycle: Concept → Customization → Impact
A clear, structured way to think about turning LLM capabilities into business outcomes.

If you are building production AI systems, leading an LLM program, or preparing for the next wave of enterprise adoption, you should definitely get a copy of this book.

Enterprise AI is evolving fast. Understanding these design patterns early puts at a career advantage which allows you to shape the next generation of intelligent applications.

#LLM

Check out this simple breakdown of the four core LLM architectures shaping most of today’s advanced systems.

Not all LLMs reason the same way. Building reliable AI systems, or choosing the right model for a workflow requires a clear mental model of the architectures that drive them.

Let’s review:

1.🔸Decoder Only (GPT, LLaMA, Grok)
Generates text by predicting one token at a time using previous context.

Best for generative workloads, conversational systems, and multi step reasoning.

2.🔸Encoder Only (BERT, RoBERTa)
Processes text with full bidirectional context to learn dense semantic representations.

Best for classification tasks, retrieval, and embedding generation.

3.🔸Encoder Decoder (T5, FLAN T5, BART)
Separates understanding from generation. The encoder builds meaning, the decoder transforms it.

Best for translation, summarization, and any structured text to text transformation.

4.🔸Mixture of Experts (Mixtral, Gemini 1.5)
Routes each input to a subset of specialized expert networks.

Best for scaling model capacity while controlling compute cost.

These architectures go beyond academic details to determine latency profiles, context handling, fine tuning strategies, and how models behave under real workloads. A clear grasp of them helps you design better pipelines, choose the right foundation models, and understand why two LLMs can produce very different outcomes on the same task.

Let me know your thoughts!

#LLM

Agentic AI vs Traditional AI: The Shift from Supervision to Autonomy

For years, Traditional AI has been like a skilled worker. It has been efficient, but dependent on human supervision.

Agentic AI, however, takes a different approach. It is a decision-maker that thinks, plans, and adapts without constant direction.

Let’s dive into what truly separates them 👇

Agentic AI: Smarter. Adaptive. Independent.

- Dynamic Learning: Perceives its environment, plans actions, and evaluates outcomes continuously.

🔸Goal-Oriented Thinking: Sets goals, creates sub-goals, and adjusts strategies like a human problem solver.

🔸Reasoning & Reflection: Learns from mistakes, refines decisions, and improves performance over time.

🔸Collaborative Agents: Multiple agents share context, delegate tasks, and synchronize for collective intelligence.

Traditional AI: Powerful but Limited.

🔹Static Learning Models: Train once, deploy, and manually retrain, with no real-time adaptability.

🔹Human-Dependent Execution: Waits for prompts or tasks, performs inference, and stops.

🔹Narrow Focus: Optimized for specific use cases without flexibility.

🔹No Self-Improvement: Requires human supervision for updates and tuning.

In Short:

- Traditional AI executes.

- Agentic AI thinks, plans, and evolves.

It’s the difference between automation and autonomy, a shift that’s redefining how AI systems operate, learn, and collaborate in the real world.

What other diffenreces have you noticed?

#AI

taplio

The Battle

Aishwarya Srinivasan

Greg Coquillo

Followers

Posts

Engagements

Likes

Performance ScoreAishwarya Srinivasan's LinkedIn Performance

Performance ScoreGreg Coquillo's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Aishwarya Srinivasan

Greg Coquillo

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Aishwarya Srinivasan's LinkedIn Performance

Performance ScoreGreg Coquillo's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAishwarya Srinivasan's LinkedIn Performance