Aishwarya Srinivasan's Linkedin Analytics

AI is obviously going to change the tech job market! Tech roles are being restructured fast: - Software engineering down 35% - AI/ML roles up 70% - Interviews are tougher - Expectations are higher If you want to stay competitive in 2025, here’s what you need to do: 1. Master the basics – Python, SQL, and core ML. Get this right first. 2. Specialize smart – Pick a lane (NLP, GenAI, CV) and go deep. 3. Don’t chase buzzwords – Focus on goals, not hype. 4. Build real projects – Show your skills in action. 5. Learn deployment – ML Ops, APIs, cloud tools = real-world impact. 6. Sharpen soft skills – Communication + problem-solving = your edge.

Last week at AI User Conference Developer Day, I had the chance to join Hina Chauhan Dixit , Subha Tatavarti and David Kanter for a deep discussion on the future of decentralized AI and running models at the edge. Here are a few highlights I walked away with: ➡️ Open source models are not just catching up, they’re outperforming! From 7B checkpoints fine-tuned for niche domains to reproducible weights with full transparency, open source is unlocking a new level of enterprise-grade control. ➡️ Enterprise use cases are shifting toward edge-first designs Latency, compliance, and data privacy aren’t optional anymore. With long-context models and edge accelerators, companies are deploying LLMs closer to the user, without compromising speed or control. ➡️ It’s not big vs. small models- it’s about the right model for the task - Personal wellness on your smartwatch? 2B models, fully on-device - Internal tooling with product knowledge? Mid-sized SLMS + RAG - Complex reasoning across domains? Cloud LLMs with privacy layers ➡️ The compute story is flipping Right now, most of the spend is on training. But as more companies move past model building and into deployment, inference will become the dominant cost, and that’s where decentralization really shines. ➡️ Privacy isn’t a feature, it’s a foundation Consumer apps want trust. Enterprises need compliance. In both cases, edge inference gives teams control over where and how data is processed. Big thanks to the organizers and everyone who joined the discussion, it’s clear that the next wave of LLM innovation is happening at the edge. PS: Thanks Kevin Mei for the amazing pictures 😄 Curious to hear: how is your team thinking about decentralization and on-device inference?

Two hours. One idea. A fully functional stock market analysis app. Sound impossible? Not anymore. Arvind just built exactly that, from scratch in less time than 2 hours. What would normally take days became achievable within mere hours, thanks to the incredible power of AI. Here’s how he did it: → App Development: Claude Sonnet 3.7 + Cursor (~1.5 hrs, with AI generating 99% of the code!) → Financial Metrics Research: Perplexity, Grok 3 (15 mins) → Sample Data Creation: Grok 3 (2 mins) → Demo Recording & Editing: Loom (15 mins) → Writing & Sharing the Story: Docs + Gemini Flash 2.0 (5 mins) AI isn’t just enhancing productivity—it’s redefining what’s possible. Check out the video demo below 👇 If you’re excited about exploring how AI can amplify your productivity, you’re in luck! We’re hosting a Vibe Coding Workshop on April 26th. Join us for an interactive session packed with live demonstrations, hands-on activities, and plenty of good vibes. Sign up today and let’s code smarter, together: https://lnkd.in/dC7aaEaN

Google just launched Agent2Agent (A2A) protocol that could quietly reshape how AI systems work together. If you’ve been watching the agent space, you know we’re headed toward a future where agents don’t just respond to prompts. They talk to each other, coordinate, and get things done across platforms. Until now, that kind of multi-agent collaboration has been messy, custom, and hard to scale. A2A is Google’s attempt to fix that. It’s an open standard for letting AI agents communicate across tools, companies, and systems, that securely, asynchronously, and with real-world use cases in mind. What I like about it: - It’s designed for agent-native workflows (no shared memory or tight coupling) - It builds on standards devs already know: HTTP, SSE, JSON-RPC - It supports long-running tasks and real-time updates - Security is baked in from the start - It works across modalities- text, audio, even video But here’s what’s important to understand: A2A is not the same as MCP (Model Context Protocol). They solve different problems. - MCP is about giving a single model everything it needs- context, tools, memory, to do its job well. - A2A is about multiple agents working together. It’s the messaging layer that lets them collaborate, delegate, and orchestrate. Think of MCP as helping one smart model think clearly. A2A helps a team of agents work together, without chaos. Now, A2A is ambitious. It’s not lightweight, and I don’t expect startups to adopt it overnight. This feels built with large enterprise systems in mind, teams building internal networks of agents that need to collaborate securely and reliably. But that’s exactly why it matters. If agents are going to move beyond “cool demo” territory, they need real infrastructure. Protocols like this aren’t flashy, but they’re what make the next era of AI possible. The TL;DR: We’re heading into an agent-first world, and that world needs better pipes. A2A is one of the first serious attempts to build them. Excited to see how this evolves.

💥Latest issue of "AI comics with Aish" is out! - Understanding Quantum Computing. If you are new to what Quantum Computing is, and how it differs from Classical Computing that uses bits (0 and 1), this is for you 👇 I love building these comics with Professor Py and Bit, who help you learn AI concepts in a fun way ❤️ PS: Thanks to GPT4o image generation, ai could build the issue 3 of the comic 8x faster than how much time it used to take me otherwise. PPS: I am always looking for new topics for the "AI Comics with Aish". Please share topics you want me to create the next issue of AI comics on, in the comments below 👇 ---------- If you enjoyed reading AI Comics with Aish, share it with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI news, educational content, and non-hype AI insights ❤️

Jensen LOVED my book “What’s Your Worth?” 👀 Who better to endorse a book on personal brand than the "Taylor Swift of Tech" himself? Okay okay… this is a ChatGPT4o-generated photo — it blows my neural net that it took me just a few seconds to create it. 🤯 In 2025, your personal brand is your resume! And that's exactly why I wrote this book to help all of you build it out strategically 💙 📣 P.S. The book is coming SOON to India and the Indian subcontinent very soon! 🇮🇳 https://lnkd.in/dv-RHX4R For all other countries, Grab your copy 👉 whatsyourworthbook.com #AI #PersonalBranding #JensenApproved (JK) #What’sYourWorth #AIhumor #IndiaLaunchComingSoon

Don't be THAT guy 🤣 If you want to code faster, smarter, and with the help of AI, then read this 👇 If you’re a student trying to build like a pro, or a founder with more ideas than time, or a software engineer ready to level up your productivity- This 3-hour AI-assisted coding workshop is built just for you! Super excited to be partnering with Replit and teaming up with Arvind Narayanamurthy to host "Vibe Coding Workshop" Here’s what we’ll walk you through (hands-on): → Rapid prototyping with AI to validate ideas → Advanced coding strategies to shorten your learning curve → Smart AI integrations to boost your workflow productivity → 2 hands-on coding examples using Replit, and Cursor You just need basic programming skills (Python preferred) and a laptop. We’ll handle the rest. Date: April 26 Time: 7 PM PST Link to join: https://lnkd.in/dC7aaEaN Seats are limited — save yours now. Let’s build the future with AI, together. P.S. Tag a friend who’s been curious about AI coding. This is their moment.

If you are starting off you ML journey in 2025, here is all you need 👇 Here are the top 10 ML open-source tools for 2025: 1. TensorFlow /Keras : ↳ A powerful and user-friendly libraries for building and training models   ↳ Great for deep learning and neural networks  💻 TF examples: https://lnkd.in/dS5w3jSb 💻 Keras examples: https://keras.io/examples/ 2. scikit-learn:  ↳ Ideal for beginners in machine learning   ↳ Offers simple and efficient tools for data mining   ↳ Supports various algorithms for classification and regression 💻 Scikit-learn examples: https://lnkd.in/dZcrsTnk 3. PyTorch/ PyTorch Lightning :   ↳ A flexible framework that supports dynamic computation   ↳ Great for research and production ↳ PyTorch Lightning simplifies the training process 💻 PyTorch Examples: https://lnkd.in/d5hkPxFe 💻 PyTorch Lightning Examples: https://lnkd.in/d8tP7978 4. XGBoost/LightGBM/CatBoost: ↳ Popular gradient boosting frameworks, excellent for tabular data tasks ↳ Highly efficient, fast training, and great predictive accuracy ↳ Used extensively in Kaggle competitions and industry for quick and reliable modeling 💻 Quick tutorial: https://lnkd.in/dR-xBW2K 5. Hugging Face Transformers: ↳ Essential for NLP and Generative AI tasks, and rapidly prototyping. ↳ Provides easy-to-use pre-trained transformer models (BERT, GPT, T5, etc.). 💻 HFT Examples: https://lnkd.in/dZABXrRK 6. Ray/ Ray Tune: ↳ Excellent for distributed computing and hyperparameter tuning. ↳ Simplifies scaling and parallelizing ML workloads. 💻 Ray Examples: https://lnkd.in/d9zwg5uZ 7. LangChain:   ↳ Perfect for building applications with LLMs  ↳ Connects LLMs with external data sources  💻 LangChain Cookbook: https://lnkd.in/d82-dKmh 8. OpenCV:   ↳ Essential for computer vision & image processing  ↳ Widely used in real-time applications 💻 Examples: https://lnkd.in/dwiSxAY7 9. MLflow ↳ Essential for model tracking, versioning, and deployment ↳ Supports reproducibility by logging experiments, parameters, and metrics 💻 MLFlow Recipes: https://lnkd.in/dtNu-g5b 10. Streamlit: ↳ Ideal for rapidly building interactive ML demos and deploying web-based applications ↳ Beginner-friendly, requires minimal frontend experience 💻 Streamlit Cookbook: https://lnkd.in/djMkEwqf 🍎 Recommendations on how to get started in comments below (I ran out of character count 😅) -------- If you found this useful, please reshare ♻️ Follow me (Aishwarya Srinivasan) for more AI news, insights, and educational content.

In 2025, your personal brand is your resume. Who are you beyond your employer, university, or other affiliations? That’s the essence of your personal brand. My personal brand? I’m a data scientist at heart, an educator passionate about making complex concepts accessible, a startup investor supporting innovation, and a dedicated health & fitness fanatic. This is who I am—regardless of the company I work for. If you still think opportunities come solely from prestigious degrees, big-name employers, or influential connections, let’s burst that myth now. Real opportunities come to those who create their own stage. When I first began sharing my journey in AI, I didn’t have a clear blueprint. All I had was genuine passion and the determination to simplify complex concepts for others. I consistently showed up, shared insights, spoke at events, and genuinely connected. Then, something amazing happened: 💙 I traveled globally, speaking about AI. 💙 Became an AI advisor, partnering with Fortune 500 companies and startups alike. 💙 Developed influence and credibility beyond my 9-to-5, unlocking career-defining opportunities. 💙 Received the prestigious EB1A ‘Einstein Visa’—with just four years of dedicated work in AI. All of this unfolded because I chose to share my authentic voice and perspective. Your personal brand isn’t just your online presence; it’s your impact, the trust you’ve built, and the opportunities you’ve attracted. It’s what people say about you when you’re not in the room, as Jeff Bezos famously said. My 2 cents: Someone out there needs exactly what only you can offer, so start now. PS: This picture is from a conference where I did a keynote and so did Patrice Evra, I met him in the green-room, got a signed copy of his book, AND gave him my book. PPS: In his talk, guess what Patrice Evra spoke about- PERSONAL BRAND! ↓↓↓↓↓↓↓ I poured my heart into my book, “What’s Your Worth?”—a practical guide designed to help you find your unique voice, passion, and true value. This isn’t a one-size-fits-all guide; it’s your compass to building a meaningful personal brand. Read a free blurb here: whatsyourworthbook.com 🚀

If you've been trying to figure out MCP vs. Function Calling, let me share how I think about it. Let's start with a quick intro about what they mean: → Function Calling is like the LLM’s personal assistant—it recognizes when the user needs something specific ("Hey, what's today's weather?") and immediately knows the right person to call (the Weather API). But once it hands off the request, your application still handles the rest—scaling, hosting, securing that endpoint. → MCP (Model Context Protocol), on the other hand, is more like a shared marketplace for AI tools. Instead of hard-wiring each individual service, you expose them once via a universal standard. Any AI model or app can discover and securely use your tools without custom integrations every single time. So here is the key difference: Function Calling = Deciding WHAT and WHEN to use a tool. MCP = Standardizing HOW tools are discovered, hosted, and consumed. Think of it this way: → Function Calling says: "I need to call an Uber right now." → MCP says: "Here’s how all rideshare services can reliably show up on one platform." If you are an AI engineer or a data scientist, here is why it matters to you: → With MCP, you don’t reinvent the wheel every time you add or move a service. → It creates interoperability—your vector search or weather tool becomes plug-and-play across any AI provider, model, or system. → It simplifies governance: built-in approval mechanisms let you audit or limit requests smoothly. Here is my take: Function Calling gave us the first taste of dynamic AI agents, and MCP is the infrastructure we need as these agents scale- so we don't drown in endless custom code. If you are curious to learn more about how you can start integration MCP servers and clients in your AI agent workflows, I would highly recommend for you to attend an upcoming FREE webinar happening on 7th May by Reid Robinson, Tal Peretz, and Matt Brown: https://bit.ly/44i4I3N I wouldn't say one versus the other, these are complimentary toolkits you nee to build and ship AI agents. PS: This lovely infographic is by Femke Plantinga 💙 --------- Found this useful? Please share it with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI insights and educational content to keep you upto date in the AI space.

Google Cloud just released one of the most insight-packed reads for AI startups! I picked up Future of AI: Perspectives for Startups at Google Cloud Next, and honestly- I love that it is designed like a coffee table book, but reads like a playbook for builders. I dug up online and found a PDF copy to share it with you all (attached below👇), and I had to share some of the takeaways that stuck with me most: 🧠 Don’t measure AI by ROI- YET! AI’s job isn’t just to cut costs. It’s to enable what wasn’t possible before. Loved Arvind Jain's perspective: stop obsessing over short-term ROI. Use AI to unearth savings, sure, but then reinvest that into topline growth. Build the things you couldn’t build before. The "10x" ideas. 🏗️ Build systems, not tools Chamath Palihapitiya reframes the game: the real value creation is in software factories- systems that turn raw business needs into output code. Less about flashy features. More about durable infrastructure that adapts and scales. 🕸️ RAG and long context aren’t competing- they’re converging It’s not RAG vs context- it’s both. Smart systems use long context to hold fluent knowledge and RAG to retrieve precision. Douwe Kiela shared some interesting thoughts on this- Train them together so they understand each other. That’s what RAG 2.0 looks like. 🔍 The next moat isn’t just data- it’s decision design How you sell AI is as important as what you build. Pricing, distribution, value alignment, these aren’t afterthoughts. They’re levers. And startups that get them right will outmaneuver slower-moving incumbents. 👩‍💻 Train your team like you train your model AI success isn’t just about features, it’s about fluency. Arvind talks about designing teams who know how to wield AI. It’s not just hiring smart people, it’s changing how your org works with tech at the center. And some honest truths I loved: → Infrastructure still breaks, it is a hard problem. You’re not the only one struggling. → Products that are easy to install are just as easy to abandon. → Most agentic AI is still duct-taped together. 📌 If you’re building in AI, this book is a goldmine. No hype. Just clear, first-principles thinking from people in the trenches. -------- If you found this insightful, do share it with your network ♻️ Follow me (Aishwarya Srinivasan) for more no-BS AI insights, news, and educational content :)

If you’re serious about breaking into GenAI roles, or prepping for your next big ML interview- this is the book you need on your shelf. Just got my hands on Generative AI System Design by ByteByteGo and I genuinely think it’s one of the most thoughtfully written technical books I’ve read in a while. It also brought back many memories. Back in grad school, I leaned heavily on the original System Design book by Alex Xu while prepping for my internship interviews. That book helped me make sense of large-scale systems and played a huge role in landing both my internship and full-time offer. Funny coincidence, a few years later, I met Alex at a creator meetup and became friends. To my surprise, he invited me to be a reviewer for his second book, Machine Learning System Design. Reviewing it was truly a full-circle moment for me 💙 If you have a copy of ML System Design Interview, you'll find my name there :) Now with their newest release- Generative AI System Design by the ByteByteGo team has truly raised the bar. Here’s what makes this book special: 👉 An insider’s perspective on what interviewers are really looking for 👉 A practical 7-step framework to help you tackle GenAI system design questions 👉10 real-world interview questions- broken down in depth 👉 280+ diagrams to make even the most complex systems easy to understand Big thanks to Alex Xu and ByteByteGo team for sending over a copy- and for continuing to create such high-quality resources for the AI community. P.S. I’ll be sharing some of my favorite frameworks and diagrams from the book soon. Stay tuned 💙 ------- Share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI insights and educational content to keep you upto date in the AI space :)

If you are an aspiring Data Scientists & AI Engineers, here are 10 Portfolio Projects for you👇 1️⃣ Stock Price Prediction with GenAI Insights Use generative AI to predict stock prices, integrating market sentiment and real-time economic indicators. ↳Tools: OpenAI API, LangChain, LlamaIndex, Prophet, Darts, Hugging Face, PyTorch Lightning, Pandas, Plotly, yFinance 2️⃣ Multimodal Sentiment Analysis Analyze sentiment from diverse data types (text, audio, video) sourced from social media platforms. ↳Tools: OpenAI GPT-4 Vision, Hugging Face Transformers, CLIP, Ollama, PyTorch Lightning, TensorFlow, OpenCV, Gradio, Pillow 3️⃣ Advanced Recommendation Engine Develop a recommendation system incorporating deep learning and user-generated content analytics. ↳Tools: OpenAI API, LangChain, LlamaIndex, Pinecone, Chroma, Surprise, LightFM, TensorFlow Recommenders, Redis, Airflow 4️⃣ AI-driven Customer Segmentation Automate customer segmentation for personalized marketing in e-commerce. ↳ Tools: scikit-learn, PyCaret, PyTorch Lightning, OpenAI API, SHAP, Plotly, Tableau, Pandas 5️⃣ Real-time Fraud Detection with AI Build systems capable of detecting fraudulent activities instantly using transaction data. ↳ Tools: OpenAI API, LangChain, PyTorch Lightning, scikit-learn, XGBoost, Apache Kafka, TensorFlow, Grafana, Prometheus 6️⃣ Predictive Healthcare Analytics Forecast patient health outcomes and potential risks using historical medical data. ↳ Tools: OpenAI API, Hugging Face (ClinicalBERT), LangChain, LlamaIndex, PyTorch Lightning, scikit-learn, Pandas, FHIR API, Streamlit 7️⃣ Real-time Autonomous Image Recognition Create image recognition systems for diagnostics, security, or autonomous navigation. ↳ Tools: OpenAI GPT-4 Vision, Hugging Face ViT/DETR/YOLO, Ollama, PyTorch Lightning, TensorRT, OpenCV, Kafka, NVIDIA Jetson 8️⃣ GenAI-powered Smart Retail Experience Enhance retail customer experience through personalized interactions and inventory management. ↳ Tools: OpenAI API, LangChain, LlamaIndex, Ollama, Pinecone, Chroma, scikit-learn, FastAPI, React.js, Streamlit 9️⃣ GenAI Customer Support Assistant Develop generative AI-powered systems to automate and enhance customer support services, build RAG for fetching internal data. ↳ Tools: OpenAI API, LangChain, LlamaIndex, Ollama, Rasa, Pinecone, Chroma, FastAPI, React.js 🔟 Predictive Maintenance Systems Predict equipment failures and optimize maintenance schedules. ↳ Tools: OpenAI API, LangChain, LlamaIndex, scikit-learn, XGBoost, PyTorch Lightning, Apache Kafka, AWS IoT, Grafana, MLflow If you were to pick ONE project from this list, which one would you choose? P.S. Repost this ♻️ so more aspiring Data Scientists & AI Engineers can see it! ---------- Follow me (Aishwarya Srinivasan) for more AI insights, news, and educational resources 🔔

🔥 ICYMI, I've got you covered with the most important takeaways from Google Cloud Next 2025. Here are the most important things you need to know + some resources you can read: 🔧 Ironwood TPUs: Google’s 7th-gen Ironwood TPUs are not just an upgrade—they redefine compute for thinking AI. At 42.5 exaflops per pod and 7.2 TB/s memory bandwidth, they're optimized for memory-heavy tasks like Mixture-of-Experts (MoE) models. The Inter-Chip Interconnect (ICI) scales to 9,216 chips, dramatically reducing latency for real-time agent collaboration. With 192GB HBM per chip, Ironwood can store a 100-trillion-parameter model entirely in memory, accelerating inference. 👉 Learn more: https://lnkd.in/g_CEFWQm 🖥️ The AI Hypercomputer:  Google’s AI Hypercomputer treats AI workloads as an integrated system. By co-designing Ironwood TPUs, NVIDIA Blackwell GPUs (via A4/A4X VMs), and Pathways runtime, it achieves near-linear scaling for billion-parameter models. Its liquid-cooled design boosts efficiency by 29x, directly cutting costs and supporting sustainability goals. 👉 Learn more: https://lnkd.in/gHANw43R 🤖 Multi-Agent Systems: The Agent Development Kit (ADK) quietly revolutionizes multi-agent development by open-sourcing the framework behind Google's Agentspace. ADK's Agent2Agent protocol enables seamless collaboration across frameworks, solving interoperability issues. ADK plus MCP, well- we are going to see a lot of vibe coded cool AI agent demos soon! 👉 Checkout ADK here: https://lnkd.in/g3AaPgas 💡 Gemini 2.5: Models That "Think" Gemini 2.5 Pro introduces built-in internal reasoning via dynamic computation graphs, boosting accuracy by 40% on complex tasks. Gemini on Distributed Cloud now enables regulated industries (healthcare, finance) to run secure, compliant, and private AI workloads with its impressive 1M-token context window. 👉 Check it out here: https://lnkd.in/ghpnpRfX 🔒 Security Google addresses AI security vulnerabilities through runtime attestation (Model Armor), preventing prompt injection attacks. Combined with Confidential Computing on Distributed Cloud, it ensures sensitive workloads remain secure and private. 👉 Check it out here: https://lnkd.in/gns7sMQ5 It was a pleasure to represent Fireworks AI at the conference and deliver a session on how Fireworks AI is able to optimize generative AI inferencing to be most efficient across speed, cost, and quality. Will soon share a blog based on my session, so stay tuned 🔔 PS: Anyone checked out ADK and built something cool? Would love to hear.

I built an AI productivity toolkit in under 10 minutes, and it’s now my favorite morning ritual 🧘‍♀️ Every day, I ask Claude: “What should I prioritize today?” And it replies with a clear list of high, medium, and low-priority tasks, pulled from my Gmail, Slack, and Google Calendar. Saves me a TON of time! I used Claude Desktop + Zapier MCP to wire this up. With Zapier’s 7,000+ app integrations, and 30,000+ actions that you can choose from, building AI-powered automations like this is easier than ever. It took me JUST 7 simple steps to turn Claude into a personal AI assistant, and YEPP can do it too. If you’re curious to try it out, I followed the Claude Desktop + Zapier MCP instructions on Zapier’s site. (Note: it’s an unofficial workaround for now and might not work for everyone.) If you want an official, plug-and-play experience, I’d recommend waiting a couple of weeks for the official connector. Or try it today with Cursor + Zapier MCP, which is officially supported. 💻 Here is a quick tutorial link for both Cursor and Claude integration with Zapier. Check it out and let me know if you try it! → To build with Claude Desktop: https://lnkd.in/dAGp9AJE → To build with Cursor: https://lnkd.in/dNQzkRM9 PS: I am using this for my personal accounts! ------------ Share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI insights and educational content to keep you up-to-date in the AI space.

The Future of AI is Open-Source! 10 years ago when I started in ML, building out end-to-end ML applications would take you months, to say the least, but in 2025, going from idea to MVP to production happens in weeks, if not days. One of the biggest changes I am observing is "free access to the best tech", which is making the ML application development faster. You don't need to be working in the best-tech company to have access to these, now it is available to everyone, thanks to the open-source community!   I love this visual of the open-source AI stack by ByteByteGo. It lays out the tools/frameworks you can use (for free) and build these AI applications right on your laptop. If you are an AI engineer getting started, checkout the following tools: ↳ Frontend Technologies : Next.js, Vercel, Streamlit ↳ Embeddings and RAG Libraries : Nomic, Jina AI, Cognito, and LLMAware ↳ Backend and Model Access : FastAPI, LangChain, Netflix Metaflow, Ollama, Hugging Face ↳ Data and Retrieval : Postgres, Milvus, Weaviate, PGvector, FAISS ↳ Large Language Models: llama models, Qwen models, Gemma models, Phi models, DeepSeek models, Falcon models ↳ Vision Language Models: VisionLLM v2, Falcon 2 VLM, Qwen-VL Series, PaliGemma ↳ Speech-to-text & Text-to-speech models: OpenAI Whisper, Wav2Vec, DeepSpeech, Tacotron 2, Kokoro TTS, Spark-TTS, Fish Speech v1.5, StyleTTS (I added more models missing in the infographic) Plus, I would recommend checking out the following tools as well: ↳ Agent frameworks: CrewAI, AutoGen, SuperAGI, LangGraph ↳ Model Optimization & Deployment: vLLM, TensorRT, and LoRA methods for model fine-tuning PS: I had shared some ideas about portfolio projects you can build, in an earlier post, so if you are curious about that, check out my past post. Happy Learning 🚀  There is nothing stopping you to start building on your idea! ----------- If you found this useful, please do share it with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI educational content and insights to help you stay up-to-date in the AI space :)

If you are building large-scale GenAI applications, and want to optimize your inference for speed, cost, and quality- we are here to help 😄 My team at Fireworks AI and I will be hanging out at the MongoDB booth on Level 1 through Google Cloud Next. Come say hi 👋

Here are 8 habits I rely on (and often suggest to others) to stay ambitious without burning out: 1️⃣ Give your ambition a time limit. Think of deep work like a meeting with your future self. Block 90 minutes early in the day, silence your phone, and go all in. When the time’s up, stop, even if you’re mid-flow. Boundaries build focus and prevent work from spilling into everything. 2️⃣ Have a “minimum viable evening.” Pick one thing that helps you unplug, cooking dinner, a walk at sunset, reading to your kid, and treat it like it’s non-negotiable. That one ritual signals the end of the workday and gives your brain a clear off-switch. 3️⃣ Check your energy, not just your to-do list. Every Friday, jot down which tasks gave you energy and which drained it. After a few weeks, you’ll see patterns. Start removing or outsourcing one draining task at a time. Over time, your schedule will start to feel less like a grind. 4️⃣ Stick to two big projects. If you’re wired to chase new ideas, this one’s hard, but worth it. Limit yourself to one main focus at work and one personal goal. Everything else goes into a “not now” list you revisit monthly. Less chaos, more progress. 5️⃣ Plan for lighter weeks. Athletes don’t train hard every day, and neither should we. Once a quarter, block a week with fewer meetings, more sleep, and no extra side projects. Building in rest makes you more resilient and keeps burnout at bay. 6️⃣ Move your body, clear your head. Doesn’t have to be fancy. A short workout, a run, yoga, anything that gets your heart rate up will help you reset and stay sharp. Exercise isn’t a nice-to-have, it’s a focus tool. 7️⃣ Short naps, big reset. Around that post-lunch slump, a 10–20 minute nap can seriously recharge you, no grogginess, just a clean mental reboot. Set a timer, close your eyes, and treat it like hitting refresh. 8️⃣ Group your tasks by vibe. Instead of switching between totally different things all day, chunk your time into themes, meetings, deep work, admin, etc. Then batch similar tasks together. Your brain stays in one lane longer, which helps with momentum. Start small, try one of these this week. You don’t need to slow down your ambition to feel more in control.

I used to think the more code I wrote, the more productive I was. Let me tell you the story of how I went from being an ML engineer to becoming an AI Developer Advocate, and what I learnt in the journey 👇 Back in 2016, as ML intern at Microsoft, I measured my value in lines of code and model accuracy. If I wasn’t typing furiously, I felt like I wasn’t contributing enough. But over time, I realized something that completely changed how I approach work: It’s not about how much you build- it’s about why you’re building it. After that internship, I went on to do my Master’s in Data Science at Columbia University. I thought I was going there to learn more machine learning techniques, and I did- but what I really learned was how to think critically, ask better questions, communicate complexity simply, and work with a team of varied skills. Getting deep into applied machine learning set me up for roles at IBM and Google as a Data Scientist. I was surrounded by brilliant minds, solving hard AI problems- but again, the biggest lesson wasn’t technical. It was learning how to make your work matter in the context of the business, it was about managing up, it was about getting stakeholder agreement. While I was deep into building ML models, my love for startups and investing in the best founders out there, made me want to try something different, and that's when I decided to take a leap of faith, I jumped into a very unique role at Microsoft for Startups as a Senior AI Advisor. It was one of the best exposures I have had- working with startups, incubators and VCs. Well, I guess I loved the startup ecosystem so much that I had to be closer to it, and go all in on not just being an ML professional but to get my hands dirty on wearing multiple hats, and that made me join Fireworks AI as Head of AI Developer Relations- Growth. Now my day-to-day is not just about building, but understanding market timing, user feedback, go-to-market strategy, keeping myself upto date on each and everything happening in the AI space, and of course working with the AI developer community. THIS stretched me in the best way possible. Every role and experience has taught me something different. → Engineering taught me precision and execution → Academia taught me curiosity and going deep → Big Tech taught me scale and people management → Startups taught me speed, adaptability, and business sense → and DevRel is (still) teaching me how to connect with people and build trust at scale To anyone early in your career: Don't stress about having it all figured out. Try different things. Be okay with pivoting. The best learning happens outside your comfort zone. And remember- productivity isn't about how busy you look. It's about how intentional you are :)

I’m Quitting LinkedIn—Because It’s No Longer Authentic Yes, you read that correctly. After a lot of thought, I’ve decided to completely step away from LinkedIn. Why am I leaving? ➡️ The feed is crowded with AI-generated posts and clickbait headlines. I see absolutely no authentic perspectives. ➡️ It takes me 10 min of scrolling to find 1 good piece of content which has some added value. My feed is full of rehashed AI generated content and AI generated comments. ➡️ Instead of spending hours sifting through AI spam, I want to channel my energy into meaningful projects—ones that prioritize real human connection and original thinking. ➡️ And most importantly, it's posts like this on the first day of the fourth month :) Edit- Had to add a hashtag as most people are not getting it 😂 #aprilfools