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
Contributing to open source genuinely changed the trajectory of my career.
My first open source PR was fixing a typo in the README of a popular ML library. Tiny change. But it got me comfortable with the contribution process, and I started doing more: adding…
By Pranay Mahendrakar Read postA bank I worked with was detecting fraud in daily batches.
The problem: by the time they flagged a fraudulent transaction, the money was already gone. They were always 24 hours behind the criminals. We rebuilt it as a streaming ML pipeline…
By Pranay Mahendrakar Read postThe smartest AI team I ever worked with shipped nothing for 18 months.
Four PhDs. Brilliant researchers. Could discuss the latest papers for hours. Could not deploy a model to save their lives. Meanwhile, a team down the hall — an ML engineer, a data…
By Pranay Mahendrakar Read postIf someone asked me "what's the single most important concept in modern AI?" I'd say embeddings…
Everything in modern AI gets converted to vectors. Words, images, audio, code, users, products — all represented as points in high-dimensional space where similarity equals proximity…
By Pranay Mahendrakar Read postI've delivered 250+ AI systems to clients worldwide.
The other 60%? Understanding the client's actual problem (which is usually different from what they initially describe), managing expectations, communicating progress in non-technical…
By Pranay Mahendrakar Read postAnomaly detection is the least glamorous ML application and possibly the most valuable.
Nobody writes excited LinkedIn posts about it. But it's silently saving millions of dollars every day — catching fraudulent transactions, predicting equipment failures before they…
By Pranay Mahendrakar Read postThere's a famous paper from Google called "Machine Learning: The High Interest Credit Card of…
ML technical debt is worse than software technical debt because it's invisible. Your model doesn't throw errors — it just quietly gets worse. Your features become entangled in ways…
By Pranay Mahendrakar Read postThere will be 50 billion IoT devices by 2030.
AI + IoT is the convergence that doesn't get enough attention because it's not as photogenic as chatbots. But it's arguably more impactful. Smart factories with predictive maintenance…
By Pranay Mahendrakar Read post10 AI startup ideas that are technically feasible today with existing tools but most people…
An AI legal contract analyzer that flags risky clauses and suggests alternatives. A personalized AI tutor that adapts to each student's learning pace and style. Real-time translation…
By Pranay Mahendrakar Read postI proposed a deep learning solution for a tabular data problem last year.
XGBoost won. By a meaningful margin. Trained in minutes instead of hours. Required no GPU. Was more interpretable. Used a fraction of the data. This happens more often than the deep…
By Pranay Mahendrakar Read postI teach AI at Tutorials Point.
What I've observed: AI in education works best when it amplifies teachers, not replaces them. The best applications aren't "AI teacher" — they're "AI teaching assistant." Adaptive…
By Pranay Mahendrakar Read postIf you can't explain the attention mechanism, you don't truly understand modern AI.
Here's how I explain it to someone new: imagine you're reading a long document and someone asks a question. You don't re-read the entire document equally. You ATTEND to the relevant…
By Pranay Mahendrakar Read post