Aishwarya Srinivasan's Linkedin Analytics

Happy Father’s Day, Appa. It’s been over six years since you left us, but not a single day goes by without me thinking of you, and thinking how I could share all my wildest dreams with you. You were my loudest cheerleader. The one who told me to dream beyond my limits and chase those dreams without fear. You were the first person who saw the creative spark in me, and you never tried to contain it. You encouraged every wild idea, every stubborn pursuit, and embraced every side of me with patience and love. You taught me that I was no less than anyone else, not less than any boy, not less than anyone around me. And it’s that belief, that upbringing, that gave me the confidence and boldness to carve my own path and stand tall in this world. You showed me how a man should treat a woman, with respect, dignity, and unwavering support. And you lived it every single day. I just wish you were here to see it all. The plans we spoke about are slowly unfolding, one by one. And there’s still so much more I want to do. I know you’re watching from above, cheering me on, just like you always did. Happy Father’s Day, Appa. We may not have been emotionally expressive to each other, but I miss you, and I love you. Always. I will keep striving to make you proud ❤️ P.S. I only have three or four pictures with my Appa, and this one is the most decent one I could find to share with you today.

If you’ve ever thought angel investing is only for rich people in closed-door circles, you’re not wrong! The first time I invested, it wasn’t through a syndicate or formal program. It was through friends, people I trusted who were building something meaningful and needed early believers, not big checks. That’s when it clicked: you don’t need millions to be an angel investor. You need conviction, curiosity, and the willingness to learn. Later, I joined Angel Squad by Hustle Fund, and that helped me formalize my approach, giving me access to vetted deals, shared evaluations, and a network of other investors. Here’s the 5-step roadmap that helped me start, and can help you too: 1. Start with what (and who) you know My first few investments were into startups founded by people I knew personally. They weren’t raising huge rounds, just looking for people who understood their vision. → Even a $1K–$2.5K check counts → What matters is: do you believe in the founder, and can you add value? 2. Invest in what you understand At the early stage, things change, products pivot, markets shift. → Focus on sectors where you understand the problem deeply → Ask: do I believe this founder can adapt when things get hard? In the beginning, don’t chase hype, bet on founders who are resilient and sharp. 3. Learn the basics of angel investing You don’t need an MBA, but you do need to understand: → SAFE vs Convertible Notes → Cap tables and dilution → Pre-money vs Post-money valuation → Pro-rata rights and exit paths Great resources: → Angel Investing School → YC’s SAFE Docs → AngelList Glossary → YC’s Startup School 4. Join a community and learn with others Joining Angel Squad helped me learn how experienced angels evaluate deals. You don’t have to do this alone, great communities include: → Angel Squad → On Deck Angels → AngelList Syndicates → VC Starter Kit These networks help you learn faster and access stronger deal flow. 5. Start small, stay curious, and add value You don’t need deep pockets, just thoughtful conviction. → Start with small checks you’re comfortable with → Track your decisions and ask for updates → If you can, offer your domain knowledge, advising shows you’re invested beyond capital And one final note, angel investing is a long-term game. If you’re expecting quick returns, this probably isn’t for you. These are illiquid, high-risk bets that take years to play out. But if you believe in the builders, and you’re excited to help shape early-stage ideas, there’s real joy in that. Always happy to chat or share resources if you’re exploring this path.

If you’re an AI engineer building multi-agent systems, this one’s for you. As AI applications evolve beyond single-task agents, we’re entering an era where multiple intelligent agents collaborate to solve complex, real-world problems. But success in multi-agent systems isn’t just about spinning up more agents, it’s about designing the right coordination architecture, deciding how agents talk to each other, split responsibilities, and come to shared decisions. Just like software engineers rely on design patterns, AI engineers can benefit from agent design patterns to build systems that are scalable, fault-tolerant, and easier to maintain. Here are 7 foundational patterns I believe every AI practitioner should understand: → 𝗣𝗮𝗿𝗮𝗹𝗹𝗲𝗹 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Run agents independently on different subtasks. This increases speed and reduces bottlenecks, ideal for parallelized search, ensemble predictions, or document classification at scale. → 𝗦𝗲𝗾𝘂𝗲𝗻𝘁𝗶𝗮𝗹 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Chain agents so the output of one becomes the input of the next. Works well for multi-step reasoning, document workflows, or approval pipelines. → 𝗟𝗼𝗼𝗽 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Enable feedback between agents for iterative refinement. Think of use cases like model evaluation, coding agents testing each other, or closed-loop optimization. → 𝗥𝗼𝘂𝘁𝗲𝗿 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Use a central controller to direct tasks to the right agent(s) based on input. Helpful when agents have specialized roles (e.g., image vs. text processors) and dynamic routing is needed. → 𝗔𝗴𝗴𝗿𝗲𝗴𝗮𝘁𝗼𝗿 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Merge outputs from multiple agents into a single result. Useful for ranking, voting, consensus-building, or when synthesizing diverse perspectives. → 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 (𝗛𝗼𝗿𝗶𝘇𝗼𝗻𝘁𝗮𝗹) 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Allow all agents to communicate freely in a many-to-many fashion. Enables collaborative systems like swarm robotics or autonomous fleets. ✔️ Pros: Resilient and decentralized ⚠️ Cons: Can introduce redundancy and increase communication overhead → 𝗛𝗶𝗲𝗿𝗮𝗿𝗰𝗵𝗶𝗰𝗮𝗹 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 Structure agents in a supervisory tree. Higher-level agents delegate tasks and oversee execution. Useful for managing complexity in large agent teams. ✔️ Pros: Clear roles and top-down coordination ⚠️ Cons: Risk of bottlenecks or failure at the top node These patterns aren’t mutually exclusive. In fact, most robust systems combine multiple strategies. You might use a router to assign tasks, parallel execution to speed up processing, and a loop for refinement, all in the same system. Visual inspiration: Weaviate ------------ If you found this insightful, share this with your network Follow me (Aishwarya Srinivasan) for more AI insights, educational content, and data & career path.

If you are building AI agents or learning about them, then you should keep these best practices in mind 👇 Building agentic systems isn’t just about chaining prompts anymore, it’s about designing robust, interpretable, and production-grade systems that interact with tools, humans, and other agents in complex environments. Here are 10 essential design principles you need to know: ➡️ Modular Architectures Separate planning, reasoning, perception, and actuation. This makes your agents more interpretable and easier to debug. Think planner-executor separation in LangGraph or CogAgent-style designs. ➡️ Tool-Use APIs via MCP or Open Function Calling Adopt the Model Context Protocol (MCP) or OpenAI’s Function Calling to interface safely with external tools. These standard interfaces provide strong typing, parameter validation, and consistent execution behavior. ➡️ Long-Term & Working Memory Memory is non-optional for non-trivial agents. Use hybrid memory stacks, vector search tools like MemGPT or Marqo for retrieval, combined with structured memory systems like LlamaIndex agents for factual consistency. ➡️ Reflection & Self-Critique Loops Implement agent self-evaluation using ReAct, Reflexion, or emerging techniques like Voyager-style curriculum refinement. Reflection improves reasoning and helps correct hallucinated chains of thought. ➡️ Planning with Hierarchies Use hierarchical planning: a high-level planner for task decomposition and a low-level executor to interact with tools. This improves reusability and modularity, especially in multi-step or multi-modal workflows. ➡️ Multi-Agent Collaboration Use protocols like AutoGen, A2A, or ChatDev to support agent-to-agent negotiation, subtask allocation, and cooperative planning. This is foundational for open-ended workflows and enterprise-scale orchestration. ➡️ Simulation + Eval Harnesses Always test in simulation. Use benchmarks like ToolBench, SWE-agent, or AgentBoard to validate agent performance before production. This minimizes surprises and surfaces regressions early. ➡️ Safety & Alignment Layers Don’t ship agents without guardrails. Use tools like Llama Guard v4, Prompt Shield, and role-based access controls. Add structured rate-limiting to prevent overuse or sensitive tool invocation. ➡️ Cost-Aware Agent Execution Implement token budgeting, step count tracking, and execution metrics. Especially in multi-agent settings, costs can grow exponentially if unbounded. ➡️ Human-in-the-Loop Orchestration Always have an escalation path. Add override triggers, fallback LLMs, or route to human-in-the-loop for edge cases and critical decision points. This protects quality and trust. PS: If you are interested to learn more about AI Agents and MCP, join the hands-on workshop, I am hosting on 31st May: https://lnkd.in/dWyiN89z If you found this insightful, share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI insights and educational content.

Finally met in person with Deedy Das last week! We have had interesting conversations over LinkedIn about how the AI startup ecosystem is evolving, and it was fun to meet in person and not just nerd out about all things AI, but also share our journey from India, career journey, going from big tech to working at a startup! Deedy is one of my fav people to learn about AI, startup ecosystem and VCs, and if you would like to get no-BS insights about everything happening around this space- definitely follow him!

If you’re building LLM applications today, reasoning is where the real leverage lies. And yet, I see a lot of engineers still treating LLM outputs as a single-shot black box. LLMs can reason, but only if you give them the right scaffolding and the right post-training. Here’s a mental model I’ve been using to think about LLM reasoning methods (see chart below): ✅ Inference-time reasoning methods: These are techniques that can be applied at inference time, without needing to retrain your model: → Tree of Thoughts (ToT), search through reasoning paths → Chain of Thought (CoT) prompting, prompt models to generate intermediate reasoning steps → Reasoning + Acting, use tools or function calls during reasoning → Self-feedback, prompt the model to critique and refine its own output → Episodic Memory Agents, maintain a memory buffer to improve multi-step reasoning → Self-consistency, sample multiple reasoning paths and select the most consistent answer ✅ Training-time enhancements: Where things get really powerful is when you post-train your model to improve reasoning, using human annotation or policy optimization: → Use Preference pairs and Reward Models to tune for better reasoning (RFT, Proximal PO, KL Regularization) → Apply RLHF, PPO + KL, Rejection Sampling + SFT, Advantage Estimation, and other advanced techniques to guide the model’s policy → Leverage multiple paths, offline trajectories, and expert demonstrations to expose the model to rich reasoning signals during training Here are my 2 cents 🫰 : If you want production-grade LLM reasoning, you’ll need both, → Smart inference-time scaffolds to boost reasoning without slowing latency too much → Carefully tuned post-training loops to align the model’s policy with high-quality reasoning patterns → We’re also seeing increasing use of Direct Preference Optimization (DPO) and reference-free grading to further improve reasoning quality and stability. I’m seeing more and more teams combine both strategies, and the gap between "vanilla prompting" and "optimized reasoning loops" is only getting wider. ---------- Share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI educational content! Image inspo: LLM Post-Training: A Deep Dive into Reasoning Large Language Models paper

Just got back from Snowflake Summit, and it sparked some important reflections on where AI is really headed in production. One theme came through loud and clear: AI is no longer a side project, it is becoming part of the operational backbone of modern enterprises. A few shifts I noticed: → From data-first to AI-first thinking We're seeing AI being used upstream - to prepare, contextualize, and govern data before it drives downstream actions. Semantic layers and agent orchestration are now essential tools. → Agentic AI is moving past the hype cycle The real conversations were about architecture - memory management, feedback loops, and designing for safety. It is encouraging to see the focus moving to engineering grounded, reliable agents, not just demos. → Governance is front and center Lineage, explainability, and trust must be integral at every layer. Building GenAI systems without this foundation will not scale safely. What stood out most was Snowflake’s pragmatic approach - building the rails for robust, governed GenAI, grounded in real-world enterprise needs. Some of my favorite moments: ❄️ Seeing Canva enhance creativity at scale with AI ❄️ Watching multi-modal complaint data processed live in Cortex ❄️ Hearing business users ask fuzzy questions, and get explainable answers in Snowflake Intelligence This Summit reinforced a core truth: it is not about chasing model benchmarks or flashy UX alone. It is about building AI that actually works - reliably, safely, and at scale. #SnowflakePartner #SnowflakeSummit

Do you have one of those people in your life- who have impacted your life/career like magic! Armand has been one of those people for me. I joined his data science team at IBM in 2019 as an intern, and he has always been incredibly supportive, been a mentor, an ally, someone I now call a friend, and can talk to for hours about a ton of common interests. Whether it was trusting me for critical projects, or being my cheerleader, or intently positioning me for my next level in career- he was always there. This is what makes someone a true leader! Today, I delivered a keynote at TechEx AI & Big Data Expo today, talking about accelerating agentic AI systems in real-world workflows. But what made it even special was sharing the stage with Armand Ruiz. I am grateful to have gotten a chance to work with him and continue to learn from him! If you are an AI professional or passionate about the space, Armand has to be the one you should follow! PS: The picture on the left is from today and the picture on the right is from 2019 from a team event at IBM ❤️

I didn’t wait for college to teach me how the real world works. I built my own curriculum, one internship at a time. During my undergrad in India, I did 11 internships. Between semesters, during semesters, whenever I could, I was out there learning by doing. Why? Because I knew early on that university degrees alone wouldn’t prepare me for how the world really works. Not all of my internships were in the same space. Some were in machine learning and data analytics. Others? → Financial forecasting → Building chatbots → Edtech product work → Prescriptive analytics pipelines → Socio-economic research at a management school Each one gave me something new, A new tool, a new mindset, a new perspective. I wasn’t optimizing for a “perfect resume”, I was optimizing for range. And the truth is, your skills are your superpower. No degree, no job title, no fancy company can define that for you. Your skillset is invincible. It grows with you, it stays with you. And that is the foundation of your personal brand. Your personal brand isn’t just what you’ve done, It’s what you can do. It’s how you show up, how you think, how you solve problems. It’s how people experience you. So if you’re starting your first internship, or even your fifth, here’s what I’d say: 🧠 Go in with a mindset to build, not just check boxes. Don’t chase titles or certificates. Chase skills. 📣 Ask all the questions. Especially the ones that make you feel silly. They spark conversations. They show you care. 🔁 Experiment fearlessly. Try different domains. Stretch your thinking. 💬 Be proactive. Set up 1:1s. Say hi. Learn what other teams are doing. At IBM, I did this religiously, and some of those connections still last today. 🌱 Never feel small in a room. Every VP once had a first internship. Don’t let titles intimidate you, let them inspire you. And remember this, You are not your degree. You are your skills. Your personal brand is not your job title. It’s how you grow, how you give, and how you show up. Happy learning ❤️ I’ve shared more of my personal journey and mindset shifts like this in my book, you can check it out here if you're curious: whatsyourworthbook.com

Headed to Amsterdam 🇳🇱 I’ll be speaking at TNW Conference this week on a topic that’s super close to my heart, Beyond the Model: Building Scalable Agentic Systems with Open Source AI. Right now, the conversation in AI is moving fast, it’s not just about building bigger models anymore. It’s about building real-world, production-grade AI systems that can reason, act, and drive business value. In my talk, I’ll be sharing, → Why agentic AI is only as good as its production performance → How to combine open-weight models with the right inference infrastructure like Fireworks AI, tool protocols (like MCP), and orchestration layers → What I’m seeing work in the field today, from architecture patterns to infra choices If you’re building in open-source AI, I’d love to see you there. And, if you’re in Amsterdam (whether attending TNW or just in town), I’m also hosting a casual AI Coffee Meetup ☕ at STRAAT Café (walking distance from the venue) on June 19th at 3PM CEST. No agenda, no talks, just good coffee and great conversations with fellow AI engineers, builders, researchers, and anyone curious about this space. If you’d like to join, please fill out this quick form so I know who’s coming 👉 https://lnkd.in/d25yTNkq

Most people overestimate what they can do in a week… But they dramatically underestimate what a decade of focused self-investment can do. Ten years. Of showing up. Of choosing growth over comfort. Of doing the quiet work when no one’s watching. That’s the difference between being who you are, and becoming who you’re meant to be. Here’s what I’ve learned: 🧠 Your brain is still becoming. Use that. The part of your brain responsible for strategy, foresight, and discipline, the prefrontal cortex, keeps developing well into your twenties. But here’s the truth: it keeps evolving with intentional effort. You are not stuck. You’re not fixed. The more you teach your mind discipline, focus, and clarity… the better it performs for you. 📖 Your habits are writing your story Every single day, you’re casting a vote for the kind of person you want to become. Not through massive breakthroughs, but through small, consistent behaviors. The way you spend your mornings. The thoughts you feed. The commitments you keep, especially the ones no one else sees. This is where identity is forged. Not in ambition, but in discipline. 🔍 Your self-awareness is your power When you get quiet enough to observe your own patterns, your emotional triggers, the beliefs you inherited, the stories you’re repeating, you give yourself a choice: 👈 To react from your past, or to respond from your future 👉 This kind of reflection isn’t soft work. It’s the hardest, most necessary work. 🚀 You’re not a victim of life. You’re the designer of it. Everything changes the moment you realize: You’re not waiting to be chosen. You’re choosing yourself, over and over again- and that choice COMPOUNDS! 🎯 Audacious goals aren’t unrealistic. They’re under-scheduled. Your dream life needs structure. Visualize it clearly. Then break it down. Not into to-do lists, but into systems. Build feedback loops. Track tiny wins. Reward effort. Iterate quickly. You don’t need to leap. You just need to keep moving- consistently, intentionally, daily. 🌱 And you can’t do it alone. Your environment is not neutral. Mentors, partners, and people with original minds will either accelerate or decelerate your growth. You need mirrors, the kind that reflect your potential, not your fears. Find them. Keep them close. If you give one decade your full self, your effort, your curiosity, your honesty, your resilience, trust me → It will give you back a version of yourself you couldn’t have imagined at the start. That’s the kind of ROI no investment beats. I’ve shared more of my personal journey and mindset shifts like this in my book, you can check it out here if you're curious: whatsyourworthbook.com

It’s been over 10 years since I started writing here on LinkedIn. Most of what I’ve shared here has been about what’s recent in AI, and more intermediate to advanced topics, because that’s the space I live and work in day-to-day. But over the past years, I’ve been getting a lot of questions from folks, especially students, early-career professionals, and those just starting out in AI, asking me: “How did you build your career?” “Where should I start if I want to get into AI?” “How do you approach personal branding in tech?” “What keeps you motivated through the ups and downs of this journey?” While I try to be as responsive as I can on LinkedIn messages, the platform is not optimized for community interactions. Plus, didn’t want to oversaturate my LinkedIn feed with too much beginner content, so I’ve been building out a space on Instagram where I share a lot more bite-sized AI learning, career tips, personal branding tips, how-to tutorials, and my lessons learnt working in the AI field, that can help you get started and grow in this field. I try my best to explain things in a way that’s comprehensive and easy to understand. One thing I’ve always had a knack for is storytelling, and that’s what I’m leaning into as I teach and share more AI concepts now. ❤️If you haven’t been following me there yet, I’d love for you to join: https://lnkd.in/denE_Zpw And beyond the AI content, I share a lot of my own journey there: ✨ Lessons I’ve learned along the way 📚 Personal stories and reflections 🎥 A peek into my daily life in AI 💭 Motivational reminders that help me stay grounded and focused It’s also a much more interactive space. I ask questions, run polls, and use my Stories to have real conversations with the community. My goal is to make it a true two-way street, a place where we can have an open dialogue, where you can ask me anything about AI, my career journey, or just life in general. It’s a space for unfiltered thoughts, and I love answering your questions there. If that sounds helpful, come join the journey, I’d love to have you there. 🚀

Proud to be back on a Times Square billboard featured again by topmate.io, continuing a tradition that’s meant a lot to me. It’s been almost three years since I started mentoring on Topmate, and I’ve truly enjoyed every part of the journey. Helping others navigate the same questions I once had has been one of the most fulfilling parts of my career. Over the years, I’ve had thousands of conversations with people around: → Breaking into data science and AI careers → Navigating EB-1A and international student journeys → Building a strong personal brand as a tech professional → Transitioning into AI from non-coding or non-technical backgrounds Whether it’s figuring out your first steps or scaling to the next stage, I’ve always believed in paying it forward. Nothing brings me more joy than hearing from someone that they finally got that job, that visa, or that opportunity,and knowing I had even a small role in their journey. Their wins feel personal to me, and that’s where I find the deepest sense of purpose. And that’s why this moment feels extra special, not just as a mentor, but as a proud investor in Topmate. It’s been incredible to watch them grow, from a small idea into a powerful global network of mentors and learners. Their pace of execution, their vision, and the community they’ve built is inspiring. Kudos Ankit and Dinesh for this amazing platform 🙏❤️ If you’re building your career in AI or data, just know: you don’t have to do it alone. You can grab a time with me: https://lnkd.in/dY9Wr8gf Let’s keep helping each other rise. PS: I am also hosting a workshop on MCP coming Sat (31st May): https://lnkd.in/dWyiN89z

Some days, I get messages from people who say: “I have so many ideas, side projects, business plans, content I want to share, things I want to build, but I don’t know where to start. How do you manage to do it all?” The truth is, I’ve asked myself that same question many times. Balancing a full-time job, being a creator, advising startups, writing a book, and still having the energy to pursue personal projects isn’t easy. It’s not about being a superhero. It’s about building systems that work with you, not against you. If you’re sitting on ambition and asking “Where do I begin?”, here’s what’s helped me move from thinking to doing: 🍋 Prioritize like your time depends on it Every idea sounds exciting. But not every idea needs your attention right now. I use the Eisenhower Matrix to separate what’s urgent, what’s important, and what can wait. The goal isn’t to do everything. It’s to do the right things first. 🍋 Make your calendar your co-founder I time-block religiously. Content planning, calls, deep work, reflection, they all have a slot. Otherwise, they don’t happen. Treat your calendar like a blueprint for your ambition. 🍋 Give yourself structure, not pressure I have dedicated days for writing, for meetings, for creative flow. You don’t need to “do it all” every single day. You just need a rhythm you can keep up with. 🍋 Build in learning Some of the best ideas come when I’m not actively working. I let AI summarize books, listen to podcasts during walks, and use small windows of time to stay curious. 🍋 Most importantly, be kind to yourself You’re not behind. You’re not late. You’re just getting started. Every system I have now was built through trial, error, and late nights wondering if it was worth it. Spoiler, it always is. If you’re feeling stuck because your ambition feels bigger than your capacity, don’t shrink the dream, start designing a system around it. Start with one idea. One calendar block. One hour a week. And trust that momentum builds. You don’t need to have it all figured out. You just need to begin. Happy growing 💛🤗🍋 I’ve shared more of my personal journey and mindset shifts like this in my book, you can check it out here if you're curious: whatsyourworthbook.com

If I think about my mom’s career, it looked very different from ours today. She worked at the same company for 30 years. Back then, loyalty made a lot of sense: → Companies provided stability and growth in return. → Long-term employees were valued and rewarded. → Staying at one place meant deeper relationships, mentorship, and trust. But the reality has changed. When I told my mom that in our generation, it’s common to switch companies every 2-3 years, she was honestly surprised. She asked, “But why?” Here’s my personal take: Loyalty to a company only makes sense if your career growth is tied to it. If you’re learning, appreciated, rewarded, and see a path forward, great. But staying put just because? That’s not the world we’re in anymore. We’ve seen how even high-impact employees, with 10+ years at a company, can be let go overnight. And that’s the harsh reality of today’s market. This is why I strongly believe in building your own brand and your own skill set, independent of any employer or title. → You’re not obliged to a company because they “gave you” a job. → You are getting paid for the value you deliver. In many cases, you might be delivering more. → Your identity should be built on your capabilities, not your badge. Here’s what I remind myself often: If at any point you feel your team or company goals no longer align with your career goals, or if you feel stuck and not growing - don’t hesitate to move. Switch to a new team. Join a new company. Start your own thing. Don’t get trapped in the comfort zone and look back later wondering why your career didn’t go where you wanted it to. In the end - you owe it to yourself to keep learning, growing, and shaping the trajectory you want. #JustMy2Cents

Agentic AI Design Patterns are emerging as the backbone of real-world, production-grade AI systems, and this is gold from Andrew Ng Most current LLM applications are linear: prompt → output. But real-world autonomy demands more. It requires agents that can reflect, adapt, plan, and collaborate, over extended tasks and in dynamic environments. That’s where the RTPM framework comes in. It's a design blueprint for building scalable agentic systems: ➡️ Reflection ➡️ Tool-Use ➡️ Planning ➡️ Multi-Agent Collaboration Let’s unpack each one from a systems engineering perspective: 🔁 1. Reflection This is the agent’s ability to perform self-evaluation after each action. It's not just post-hoc logging—it's part of the control loop. Agents ask: → Was the subtask successful? → Did the tool/API return the expected structure or value? → Is the plan still valid given current memory state? Techniques include: → Internal scoring functions → Critic models trained on trajectory outcomes → Reasoning chains that validate step outputs Without reflection, agents remain brittle, but with it, they become self-correcting systems. 🛠 2. Tool-Use LLMs alone can’t interface with the world. Tool-use enables agents to execute code, perform retrieval, query databases, call APIs, and trigger external workflows. Tool-use design involves: → Function calling or JSON schema execution (OpenAI, Fireworks, LangChain, etc.) → Grounding outputs into structured results (e.g., SQL, Python, REST) → Chaining results into subsequent reasoning steps This is how you move from "text generators" to capability-driven agents. 📊 3. Planning Planning is the core of long-horizon task execution. Agents must: → Decompose high-level goals into atomic steps → Sequence tasks based on constraints and dependencies → Update plans reactively when intermediate states deviate Design patterns here include: → Chain-of-thought with memory rehydration → Execution DAGs or LangGraph flows → Priority queues and re-entrant agents Planning separates short-term LLM chains from persistent agentic workflows. 🤖 4. Multi-Agent Collaboration As task complexity grows, specialization becomes essential. Multi-agent systems allow modularity, separation of concerns, and distributed execution. This involves: → Specialized agents: planner, retriever, executor, validator → Communication protocols: Model Context Protocol (MCP), A2A messaging → Shared context: via centralized memory, vector DBs, or message buses This mirrors multi-threaded systems in software—except now the "threads" are intelligent and autonomous. Agentic Design ≠ monolithic LLM chains. It’s about constructing layered systems with runtime feedback, external execution, memory-aware planning, and collaborative autonomy. The RTPM framework is great for engineers to move beyond prototypes and toward robust, production-grade agent architectures.

What does it feel like to work at Fireworks AI ? Well, it feels rare, in the best way. Last week, we hosted our first-ever Developer Day, which was a huge hit, with over 150 people at the event, and an incredible line of speakers from Perplexity, Notion, Vercel, and Mercor, hosted by our cofounders Lin Qiao and Benny Yufei Chen. We had very exciting product launches from experimentation platform, to Reinforcement Fine-Tuning, to Voice Agents, and Virtual Cloud Infrastructure support. I was pleased to be able to host the event and get to hear and get to hear firsthand how some of the top AI engineering teams are building with open models, optimizing for real-world performance, and solving infrastructure challenges at scale. Fireworks AI team is one of the most densely talented teams I’ve ever had the privilege to work with. Every single day feels like a learning experience, and I don’t say that lightly. Working at a startup is always dynamic. Plans shift fast, priorities evolve, and you wear multiple hats. But what keeps it all grounded is the shared energy and passion everyone brings to the table. We’re all moving towards the same mission: "To make building generative AI applications seamless for developers, by serving the highest quality open-source models with unmatched speed, reliability, and developer experience." It’s rare to find a team that’s not only technically world-class, but also genuinely collaborative, thoughtful, and execution-driven. Walking into work every day feels energizing. You know you’re building something that matters- and doing it alongside people who care deeply about getting it right. Don't worry if you missed the Dev Day event, we put together a detailed blog with all of our announcements and sessions. Check it out here: https://lnkd.in/dMXQTrYE And yes, we’re hiring 😄 If you’re excited by GenAI infra, love building in public, and want to get your hands dirty with cutting-edge open-source models, come join us: https://lnkd.in/dFX6xVup

I had the exciting opportunity  to attend Snowflake Summit, not just as an attendee, but as a Snowflake Insider, and I walked away genuinely impressed. I’ve seen a lot of GenAI platforms over the past few years, but what Snowflake is building with Cortex AISQL and Snowflake Intelligence is on a different level. This isn’t just about flashy demos. It’s about real-world readiness. → Natural language querying across structured and unstructured data → Full traceability and source validation → A semantic layer that actually learns from usage patterns → And multimodal support, yes, even image and audio analysis Add to that Snowflake Openflow for seamless data integration, Horizon for end-to-end governance, and Gen2 Warehouses boosting performance, all baked right into the native stack. No third-party dependencies. No patchwork. As someone who’s worked with AI in production across several platforms, this felt like a real shift. It’s clear they’re building for scale, trust, and enterprise complexity. This is what it looks like when GenAI meets real infrastructure. If you want to learn more, check it out here: https://lnkd.in/djmQZfFU #SnowflakeSummit #SnowflakePartner

Just delivered my 100th session talking about AI, and I couldn’t have picked a better place for it than The Next Web Conference (TNW) in Amsterdam 🇳🇱 Had to shuffle some things to make it happen, but giving my milestone talk in one of the coolest cities in the world? 100% worth it. My keynote focused on: Beyond the Model: Building Scalable Agentic Systems with Open Source AI Here’s what I covered: – Most teams are still focused on improving models – But the real challenge is building systems that can actually do things reliably – Open source gives us the tools, but it also requires a systems-first mindset We also got into: → Moving from models to multi-agent systems → Architectures that scale with real-world complexity → Why open source is an advantage, not a compromise → The latest announcements from Fireworks AI, including experimentation platform/ Build SDK, supervised fine-tuning v2, Reinforcement fine-tuning, and voice agents Amsterdam was 🔥 and it was exciting to see so many people thinking beyond models, and into system-level innovation. If you’re curious about what we’re building at Fireworks AI, or want to jam on agentic workflows, DM me! Huge thanks to TNW for having me, and to everyone who came up to chat after. I also hosted a meetup post-talk and had some incredible conversations I’ll share in a follow-up post :)

I’ve never dreamt within my means. And I’ve never trusted anyone who told me to. I grew up in a home built in the basement of a parking lot. It was a one-bedroom space for four people- me, amma (mom), appa (dad), and paati (grandma). There was warmth, love, and laughter, but also leaking ceilings, shared beds, and constant reminders of what we couldn’t afford. No family vacations. No extracurriculars. No room to dream freely. Back then, I didn’t even know what studying abroad meant. It just wasn’t a world I had access to, or could even imagine myself in. It wasn’t until I joined VIT that things started to shift. I saw students around me prepping for GREs, researching universities in the U.S., talking about scholarships and assistantships. That was the first time I realized: this is possible. And the moment I knew it existed, I knew I had to go for it. When I got into Columbia University, I applied for my first passport. I cannot be more grateful to my professor at Columbia University who offered me an RA that not only covered my tuition but also paid me handsome stipend every month. That’s when the dream finally had a direction. People often ask, “What kept you going? What was your motivation?” It was this: → To give my mother the dream home she never had growing up. → To give my family the kind of comfort I only saw from a distance. → And to show my father, who passed away just a week after I graduated from Columbia - “Appa, I made it. Thank you for standing by me every step of the way.” That was my why, and I was willing to do whatever it took. I wasn’t just chasing a degree or a job, I was building a life I once couldn’t even picture. Being an immigrant in the U.S. gave me the platform to chase my wildest dreams, and the freedom to define success on my own terms. We’re built different. We carry grit in our bones and ambition in our breath. We build, not just for ourselves, but for every generation that comes after us. This AAPI Heritage Month, I celebrate not just my roots, but the road I’ve walked, and super grateful for what this country has enabled me to achieve for my family. I hope this story gives someone out there permission to dream a little bigger. #AAPIMonth