Sarthak Handa

There’s a new breed of GenAI Application Engineers who can build more-powerful applications faster than was possible before, thanks to generative AI. Individuals who can play this role are highly sought-after by businesses, but the job description is still coming into focus. Let me describe their key skills, as well as the sorts of interview questions I use to identify them. Skilled GenAI Application Engineers meet two primary criteria: (i) They are able to use the new AI building blocks to quickly build powerful applications. (ii) They are able to use AI assistance to carry out rapid engineering, building software systems in dramatically less time than was possible before. In addition, good product/design instincts are a significant bonus. AI building blocks. If you own a lot of copies of only a single type of Lego brick, you might be able to build some basic structures. But if you own many types of bricks, you can combine them rapidly to form complex, functional structures. Software frameworks, SDKs, and other such tools are like that. If all you know is how to call a large language model (LLM) API, that's a great start. But if you have a broad range of building block types — such as prompting techniques, agentic frameworks, evals, guardrails, RAG, voice stack, async programming, data extraction, embeddings/vectorDBs, model fine tuning, graphDB usage with LLMs, agentic browser/computer use, MCP, reasoning models, and so on — then you can create much richer combinations of building blocks. The number of powerful AI building blocks continues to grow rapidly. But as open-source contributors and businesses make more building blocks available, staying on top of what is available helps you keep on expanding what you can build. Even though new building blocks are created, many building blocks from 1 to 2 years ago (such as eval techniques or frameworks for using vectorDBs) are still very relevant today. AI-assisted coding. AI-assisted coding tools enable developers to be far more productive, and such tools are advancing rapidly. Github Copilot, first announced in 2021 (and made widely available in 2022), pioneered modern code autocompletion. But shortly after, a new breed of AI-enabled IDEs such as Cursor and Windsurf offered much better code-QA and code generation. As LLMs improved, these AI-assisted coding tools that were built on them improved as well. Now we have highly agentic coding assistants such as OpenAI’s Codex and Anthropic’s Claude Code (which I really enjoy using and find impressive in its ability to write code, test, and debug autonomously for many iterations). In the hands of skilled engineers — who don’t just “vibe code” but deeply understand AI and software architecture fundamentals and can steer a system toward a thoughtfully selected product goal — these tools make it possible to build software with unmatched speed and efficiency. [Truncated due to length limit. Full post: https://lnkd.in/gsztgv2f ]

5,055

263 Comments

India is not behind on talent. It is behind on risk appetite. Building LLMs or foundational AI infra is not about capability. It is about capital. Not just any capital, but capital that is comfortable with long timelines, unclear outcomes, and massive burn. Most of India’s funding still chases near-term traction. SaaS, fintech, consumer. Sectors where returns are easier to model. But AI infra does not play by those rules. It demands belief before proof. The Middle East launching a sovereign fund to accelerate AI through public private bets is not just money. That is intent. India will not build foundational AI by hoping startups will figure it out alone. It needs conviction at the system level. And a willingness to back uncertainty at scale.

56

10 Comments

𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝗶𝘁 𝗺𝗲𝗮𝗻 𝘁𝗼 𝘀𝗰𝗮𝗹𝗲 𝗰𝗼𝗻𝘀𝘂𝗺𝗲𝗿 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝘀 𝗶𝗻 𝗮𝗻 𝗔𝗜-𝗳𝗶𝗿𝘀𝘁 𝘄𝗼𝗿𝗹𝗱? Dived deep into this at the roundtable at Bits n Atoms, hosted by The Product Folks last night — and the conversations stuck with me. Scaling consumer products in an AI world demands thinking across multiple dimensions, and these are just some of the lenses through which our group explored it.  • 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘀𝗮𝘁𝗶𝗼𝗻 — not just smart recommendations, but designing for one and actually impacting behaviour change  • 𝗧𝗿𝘂𝘀𝘁 — especially for this new generation of AI-native products, where UX becomes critical to adoption  • 𝗖𝗼𝘀𝘁-𝗲𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲𝗻𝗲𝘀𝘀 — using AI to boost efficiency and also designing AI systems that can scale without exploding infra or model costs  • 𝗠𝗼𝗻𝗲𝘁𝗶𝘇𝗮𝘁𝗶𝗼𝗻 — whether through subscriptions or new models that make sense for AI-first experiences  • 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 & 𝗚𝗧𝗠 — this differentiation matters more than ever, in a world where building products is getting faster and cheaper everyday  • 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 — across regions, languages, devices, and modalities. If it only works for the few, does it really scale? We’re in an interesting moment — where product thinking is expanding, not just because of what AI can do, but because of the questions it’s forcing us to ask. Loved the sharp, honest, and wide-ranging perspectives from everyone Shreyanshi C. Kushal Khandelwal Ashish Jain Madhav Bhartia Tanmay Saxena Pranav Agrawal Sachin Kamkar Shashank Raghavendra Nikhil Ankan Huge shout out to Abhay Jani Aditya Mohanty Suhas Motwani for driving these thoughtfully curated and highly impactful events!

92

6 Comments

I had a fascinating conversation with Abhishek Das about how to win in AI (+ why horizontal > vertical) Abhishek is a pioneer in agentic AI (2016 PhD thesis on agents that can see, talk, and act + Meta FAIR Lab researcher) & co-founder of Yutori (Radical Ventures, Felicis, Elad Gil, Sarah Guo, Scott Belsky, Guillermo Rauch, etc) My biggest learnings: 1. Errors compound viciously in multi-step processes In complex workflows, each decision is a point for errors to multiply. 90% accuracy sounds great until you realize that over a 10-step workflow, you're failing more than succeeding. Over 50 steps, you're basically guaranteed to fail. On top, the long tail of edge cases in every process makes this worse. You can't train on every possible scenario. Instead, your agents need to detect mistakes quickly and backtrack. The effort to get there is enormous - getting from 50% → 80% → 99+% reliability scales almost linearly in effort. There's no shortcut. 2. Depth on one thing creates memorization, not generalization If you train too narrowly - on one customer, one workflow, one use case - your model learns the quirks, not patterns. You get brittle intelligence that can't transfer. Yutori initially built a food ordering agent for DoorDash. It worked beautifully. But when they tested it on other food sites - same category, similar UIs - it collapsed. The model had memorized DoorDash's quirks instead of learning how food ordering works. 3. To win in horizontal AI, pick a capability, then go wide Building in horizontal AI is as competitive as it gets. As a small startup, focus on a single, powerful capability, do it incredibly well, and apply it broadly. After their DoorDash experiment failed to generalize, Yutori scoped down to one capability: Agents that monitor the web for anything you care about - specific but broadly applicable. Training on all sites on the web across diverse UIs within that single capability made the model robust to become the best at this capability. 4. Research and product have to talk Many AI companies have big research teams. When research and product operate in isolation, you get either science projects that never ship or products built on shaky technical foundations. Ideas should flow both directions. Sometimes research unlocks a capability that product can ship. Sometimes product shapes the research roadmap. The key is visibility across the company. 5. In fast-moving spaces, plan in 3-month horizons Anything beyond that has too much variance - both in what you can build and what the world will look like. Have rough sketches for 6 to 12 months, but plan in quarters. You can't predict further out, pretending otherwise wastes everyone's time. 6. Everyone should read user feedback Customer feedback has to reach every corner of the company At Yutori, all onboarding emails come from the founder / replies land in his inbox. On top, everyone on the team, no matter how technical, rotates through reading user feedback

58

2 Comments

Another interesting post (https://lnkd.in/gycDa5w3) from Aditi Naval Kumar on how start-ups (and, for that matter, even mature enterprises) need to think about pivoting if their original strategies are not working out. While she has very good advice to offer, reading her blog reminded me about an enduring concern of mine, which is in part related to the idea of pivoting. It is the shift of the enterprise from problem-solving mode to an anything-to-survive mode. To me, the interesting problems to solve are hard systemic ones -- whether it is price discovery for agriculture produce, building derivative markets that transfer risks and help take markets closer to the Arrow-Debreu Complete Markets ideal, addressing credit market gaps which prevent even high quality micro-lenders serving remote populations from accessing the credit they need, or solving demand/supply aggregation problems. Solving such problems unlocks entire ecosystems and propels the equitable growth of the economy, but its value and potential can take time to realise. If investors refuse to take a portfolio view of their investments and don't have a vision of how and where they wish to add value to the ecosystem, there is a risk that you will get a lot more value-destroying in the name of pivoting.

27

2 Comments

🔍 Function Calls vs Model Context Protocol (MCP) in Agentic AI 🤖 Function calling brought structure to LLM interactions — enabling models to trigger tools, retrieve data, and execute tasks. But in agentic systems, where LLMs act as thinking agents rather than reactive bots, function calls alone don’t cut it. Enter: Model Context Protocol (MCP) 🧠 🔹 Function Calls = “Here’s a tool. Use it.” 🔹 MCP = “Here’s context, memory, history, toolset, persona, and environment — go accomplish a goal.” Why MCP matters: • ✅ Context-rich reasoning — gives the model structured access to user goals, memory, history, and tool capabilities. • 🔄 Multi-turn continuity — allows agents to persist state and adapt plans over long tasks. • 🛠️ Tool orchestration — lets the model choose and combine tools intelligently rather than just calling one-off APIs. • 📦 Modularity & Portability — MCP creates a standardized, modular interface for managing agent inputs and outputs — beyond raw JSON and ad hoc hacks. Think of function calls like giving an intern a list of API docs. Think of MCP like onboarding a smart assistant with access to calendars, tasks, Slack, past conversations, and team goals — and letting them figure out what to do next. Agentic AI needs structure beyond prompts. MCP is that structure — enabling autonomy, reasoning, and reusable frameworks at scale. We’re moving from “chat with AI” to “collaborate with AI agents.” #AgenticAI #MCP #LLMs #AIInfrastructure #ToolUse #AIContext #AIArchitecture #GenerativeAI #FunctionCalling

38

2 Comments

India’s AI journey must focus not just on consumption, but on creation. While global LLMs are powerful, they often miss India’s linguistic and cultural nuances. When a farmer in Punjab asks a question in Punjabi, or a small-town entrepreneur in Bihar seeks legal advice in Hindi, how effective can these models truly be? India’s diversity is vast — but do our AI models reflect this? 22 official languages and over 1,600 dialects Thinking shaped by culture, context, and nuance Data spread across vernacular platforms, regional websites, PDFs, handwritten documents, and more 𝗛𝗲𝗿𝗲’𝘀 𝘄𝗵𝗮𝘁 𝗜𝗻𝗱𝗶𝗮 𝗻𝗲𝗲𝗱𝘀:  ✅ 𝗟𝗟𝗠𝘀 𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝗶𝗻 𝗿𝗲𝗴𝗶𝗼𝗻𝗮𝗹 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀  ✅ 𝗕𝘂𝗶𝗹𝘁 𝗼𝗻 𝗜𝗻𝗱𝗶𝗰 𝗱𝗮𝘁𝗮𝘀𝗲𝘁𝘀 (𝗴𝗼𝘃𝗲𝗿𝗻𝗺𝗲𝗻𝘁 𝘀𝗰𝗵𝗲𝗺𝗲𝘀, 𝗹𝗼𝗰𝗮𝗹 𝗰𝗮𝘀𝗲 𝗹𝗮𝘄𝘀, 𝘀𝘁𝗮𝘁𝗲-𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗿𝘂𝗹𝗲𝘀, 𝗲𝘁𝗰.)  ✅ 𝗖𝘂𝗹𝘁𝘂𝗿𝗮𝗹𝗹𝘆 𝗮𝗹𝗶𝗴𝗻𝗲𝗱, 𝗰𝗼𝗻𝘁𝗲𝘅𝘁𝘂𝗮𝗹𝗹𝘆 𝗮𝘄𝗮𝗿𝗲, 𝗮𝗻𝗱 𝘀𝗼𝗰𝗶𝗮𝗹𝗹𝘆 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝗲 Imagine a Marathi teacher in Satara creating lesson plans in seconds. A shopkeeper in Kerala receiving tax advice in Malayalam. A student in Assam preparing for UPSC in Assamese. That’s real AI for Bharat. We’ve got the talent. We’ve got the data. Now, we need the intent and investment. #AIforBharat #IndianLLM #GenAI #MakeInIndia #DigitalIndia #LLM #AI #LanguageModels

39

10 Comments

We are experiencing a shift in product strategy. The market is moving from model intelligence to workflow ownership. As LLMs like GPT and Claude converge on similar performance benchmarks, the model itself is becoming less of a differentiator. Intelligence is becoming a utility. The model-centric approach is fading. In this approach, the primary source of value is the behavior of the underlying foundation model itself, optimized through benchmarks and prompt engineering. The risk is constant volatility. Foundation-model updates can change behavior, alter outputs, and break assumptions the product depends on. Products built this way are often one model release away from regression, rework, or obsolescence. The workflow-centric approach is rising. It prioritizes product–workflow fit over model performance. Fit-for-purpose models are embedded deeply into vertical workflows, enriched with domain context, state, and rules. The product’s value comes from completing real tasks, which is what ultimately makes it sticky. The most successful builders/entrepreneurs I see right now are not competing on foundational architecture. They are focused on embedding intelligence directly into real workflows where work gets done. Ultimately, you can buy intelligence via an API. But there are two things you cannot buy. The first is distribution. Not sales channels, but integration. The winners optimize for presence by embedding their product deep into existing systems, delivering intelligence exactly where the work is already happening. The second is context. This is the real moat. Deep domain specialization that explains why a customer is taking an action, not just what they are typing. It is accumulated through workflow exposure and cannot be replicated by general models. In the long run, sticky workflows will beat better benchmarks.

52

3 Comments

If you’re an AI engineer, LLM builder, or researcher looking to move beyond basic RAG pipelines and build production-ready systems, this workshop is for you. Arvind Narayanamurthy and I are hosting a 3-hour deep dive that goes far beyond retrieval. Here’s what we’ll cover: 🧠 RAG as a Memory Layer → How to enable both short-term and long-term memory for LLMs 🧠 Context Engineering → Selecting, ordering, and compressing context to optimize model performance 🧠 RAG Architectures → Deep dive into architectural patterns, trade-offs, and when to use what 🛠️ Hands-On 1: Build a Multimodal RAG App (text + images) 🛠️ Hands-On 2: Build a Graph RAG App 🗓️ Duration: 3 hours 💻 Format: Deep dive + live build + open discussion 🔗 Register here: https://lnkd.in/dww4Vguh This is Part 2 of our RAG Series: Towards Mastery. In Part 1, we laid the foundation, covering: ✔️ What RAG is and why it matters ✔️ How to build a basic pipeline with metadata filtering + hybrid retrieval ✔️ Vector DBs, embedding models, chunking strategies ✔️ Three hands-on coding sessions 📹 If you missed Part 1, the recording is available here: https://lnkd.in/gjx5WNP6 Whether you’re scaling your first RAG system or refining an existing one- this is where you level up. Let’s build together. 🔥 PS: Follow The Gen Academy for more applied AI workshops!

162

15 Comments

I was talking to my friend Parul Singh about how too much VC money can be detrimental. I have been working with pre-seed/seed/A at and with companies for the last 20+ years and I have observed first hand that the saying “constraints breeds creativity” is true. In my experience, a company with 18 months of runway gets much more done in the second nine months than the first. 🪹 Pre-seed Companies: When you are in the “idea maze” (coined by Andreessen Horowitz) or “drunken walk” having too much money does not necessarily speed up progress. Of course you may need some money for founder living expenses but having a larger team wandering through the "idea maze" together doesn't necessarily make the journey go faster. There is a certain level of founder reflection that has to happen while navigating the “idea maze” and too many distractions (people, money) makes it hard. I teach an entrepreneurship class  (LeanLaunchpad) every year. Out of the eight teams we have navigating the “idea maze”, about half will test and land on a reasonable idea in just ten weeks (including a simple MVP and even some paid POCs). Total budget: zero. With the multitude of AI tools, time to PMF is accelerating. Check out https://lnkd.in/guM7Ymn9 (AI companies revenue >10x SaaS companies). 🌱 Seed Companies: When you are going from the “idea maze” to “product market fit” and are trying to first build a product that a few customers love and second build a way to acquire these customers efficiently, having too much money can actually be counterproductive. Why? Because too much money often translates to wasting time on things that don’t matter: you are not forced to focus on the features that are absolutely critical, you can spend in non-efficient marketing methods (skipping the crafting of the product-growth loop) and you can afford to over hire or to keep the wrong people onboard. 🅰️ Series A Companies: Post product market fit, when it is time to scale, money is absolutely critical. However there is one common failure mode: the “fake Series As”, the As that get overpriced without really having reached product-market fit. Take a seed in a hot space, a small team that has grown revenue to a few $100Ks or maybe even a million in just six months. There is little data on actual revenue stickiness or whether the founder-led sales motion can be replicated efficiently. There is a lot of competition to invest in the company and they end up raising $20M+ series A.  The founder feels confident, they are stretched-thin and they hire a bunch of folks. There is pressure to scale but the machine is not built yet (the product is not quite there, they can’t figure out how to get to customers efficiently) and things unravel. Venture money is like carbs: quality over quantity (not all money is equal); take in moderation (constraints bread creativity) and if you compete in endurance sports, you might need more carbs to perform well (if you are growing, you will need the cash).

199

16 Comments

🚨 “𝙏𝙝𝙞𝙨 𝙞𝙨 𝙗𝙞𝙜𝙜𝙚𝙧 𝙩𝙝𝙖𝙣 𝙩𝙝𝙚 𝙞𝙣𝙩𝙚𝙧𝙣𝙚𝙩.” That’s how Tracxn's co-founder Abhishek Goyal describes the AI boom in his MIT Sloan Management Review India's interview. (Link to full post in the comments) Top AI minds are leaving Big Tech to build from scratch. White spaces. New industries. Entire job categories being redefined. 🎓 𝐖𝐡𝐚𝐭 𝐝𝐨𝐞𝐬 𝐭𝐡𝐢𝐬 𝐦𝐞𝐚𝐧 𝐟𝐨𝐫 𝐮𝐧𝐢𝐯𝐞𝐫𝐬𝐢𝐭𝐢𝐞𝐬? It's a once-in-a-generation wave of reinvention. It’s a 𝘧𝘰𝘳𝘦𝘴𝘪𝘨𝘩𝘵 𝘴𝘪𝘨𝘯𝘢𝘭. One that shows you exactly where education, research, and talent development need to go next. And, this is the moment to bring Tracxn 𝐢𝐧𝐬𝐢𝐝𝐞 𝐲𝐨𝐮𝐫 𝐜𝐚𝐦𝐩𝐮𝐬. Not just to teach AI, but track it, map it, and build with it. With Tracxn inside classroom, universities can - ✅ Power curriculum with real-time startup data ✅ Equip students with market-ready insight ✅ Forge research and placement pathways rooted in live innovation ✅ Forge new industry-academia partnerships Universities that act now won’t just keep up. They’ll lead. 👋 If you care about where your university is headed, as a leader, faculty, Dean, Director, Chancellor or well-wisher — Let’s talk. I’ll show you how Tracxn becomes part of your academic edge. #HigherEd #AIinEducation #UniversityInnovation #Tracxn #FutureOfWork #AcademicPartnerships #AIStartups #MITSMR #GlobalHigherEd #DeansOfInnovation

90

3 Comments

🎙️ Lessons from Building and Scaling at Paytm, Microsoft and HT Digital Had a great conversation with Sukant Kumar recently about the realities of tech leadership, system scaling, and AI's practical impact on engineering teams. What we covered: 🔹 Leadership Mindset Shifts – How moving into management roles changes the way you think about problems, people, and priorities 🔹 Real-World Scaling Challenges – Taking Paytm from 250 to 2,500 TPS: the 3 decisions that actually moved the needle (hint: it wasn't just infrastructure) 🔹 AI and Engineering Productivity – Beyond the hype—how GenAI is genuinely changing workflows and what it means for team composition and costs 🔹 Hiring and Team Dynamics – Interview red flags, what distinguishes exceptional engineers, and maintaining connection across large teams What working at three different companies taught me: Microsoft showed me enterprise-scale systems and processes. Paytm was all about fintech speed and handling transaction volumes under pressure. HT Digital revealed the unique challenges of media-tech and SEO-driven engineering. Same fundamentals—clean code, thoughtful architecture, strong teams. Completely different execution contexts. Also discussed: → SEO strategies across different city tiers → Handling AI hallucinations in code → Whether AI will tackle complex engineering problems soon → The human side of difficult decisions like layoffs Watch here: https://lnkd.in/gPG6iKeq What's been your experience with scaling systems or adopting AI in your engineering workflow? Would love to hear your perspective. 👇 #TechLeadership #SoftwareEngineering #Microsoft #Paytm #HTDigital #EngineeringManagement #AI #SystemDesign #Scaling #FinTech #MediaTech #GenerativeAI #EngineeringLeadership #SystemArchitecture

64

2 Comments