Notes from the
edge of the model.
Field notes on what actually breaks in production — agents, retrieval, evaluation, MLOps, and the career decisions nobody writes down. Longer arguments become papers; these are the rest.
353 posts
People ask for my certification roadmap.
Months 1-2 (Foundation): Andrew Ng's ML Specialization and Google's ML Crash Course (both free). Build two small projects. Months 3-4 (Cloud): Pick ONE — AWS ML Specialty OR GCP ML…
By Pranay Mahendrakar Read postLLMs have the memory of a goldfish.
This is one of the most important unsolved problems in AI, and solving it (even partially) is enormously valuable. The approaches that work in practice: RAG for external knowledge (the…
By Pranay Mahendrakar Read postHard truth from a decade in AI: the projects that succeeded weren't the ones with the best models.
I've worked with Mercedes-Benz Germany, the Indian Army, IIT Bombay, and dozens of other organizations. The technical challenges were real. But the projects that stalled? Almost always a…
By Pranay Mahendrakar Read postHere's a statistic that should make AI startup founders nervous: serving costs kill more AI…
Training an LLM is a one-time cost. Serving it to thousands of users 24/7 is an ongoing hemorrhage if you're not careful. I've seen startups spending $100/day on inference that could…
By Pranay Mahendrakar Read postEvery significant career opportunity I've had came from a side project.
The personal AI assistant I built for myself led to a client conversation. The code review bot I made as a weekend experiment turned into a consulting gig. A recipe generator I built to…
By Pranay Mahendrakar Read postDeepSeek proved something that a lot of well-funded AI labs didn't want to hear: you don't need…
Their efficient architecture choices delivered competitive performance at a fraction of the compute budget of Western labs. That's not just a technical achievement — it's a strategic…
By Pranay Mahendrakar Read postI deployed a model manually once. Configuration error brought down the service for 4 hours.
Never again. ML CI/CD is different from software CI/CD in ways that trip people up. With software, you test the code. With ML, you test the code AND the data AND the model AND their…
By Pranay Mahendrakar Read postI built precision lithography control systems for AMS-INDIA.
While everyone chases consumer AI applications, manufacturing companies are quietly deploying AI that saves them millions: visual quality inspection with computer vision, predictive…
By Pranay Mahendrakar Read postBefore Weights & Biases, my experiment tracking was a spreadsheet.
I cringe thinking about it. W&B solved a problem I didn't realize was costing me hours every week. Every experiment automatically logged. Hyperparameters, metrics, system stats — all…
By Pranay Mahendrakar Read postText-only RAG was the big thing in 2024.
Think about what real business documents look like. Technical manuals with diagrams. Financial reports with charts. Medical records with scans. Product catalogs with images. None of…
By Pranay Mahendrakar Read postEvery major opportunity in my career — every single one — came through a connection.
That's not luck. That's consistent investment in relationships. I don't think of networking as "collecting contacts." I think of it as contributing to a community and building genuine…
By Pranay Mahendrakar Read postMost ML engineers stop when the metrics look good. The best ones start there.
Systematic error analysis is the most underrated skill in machine learning. It's also the one that separates deployed models that actually work from ones that look good on dashboards but…
By Pranay Mahendrakar Read post