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
I've delivered AI systems for the Indian Army.
The systems MUST be reliable. There's no "it works 90% of the time" when lives are on the line. Every edge case matters. Every failure mode is documented and tested. Every prediction has…
By Pranay Mahendrakar Read postKaggle taught me more practical ML than my entire formal education. And I didn't even need to win.
My best learning came from consistently finishing in the top 10-20% and then reading the top solution write-ups. Those write-ups are an absolute gold mine — they show you the gap between…
By Pranay Mahendrakar Read postAI infrastructure engineering pays 30-50% more than regular ML engineering.
Here's what AI infra engineers build: GPU clusters for training (keeping thousands of GPUs running efficiently is harder than it sounds). Feature computation platforms that serve…
By Pranay Mahendrakar Read postThink of Model Context Protocol (MCP) as USB-C for AI agents.
Before MCP, connecting an AI agent to external tools meant custom integrations for each one. Want your agent to use a database? Custom code. Slack? Custom code. File system? Custom code…
By Pranay Mahendrakar Read postI have 7 published research papers, 3 books, a UK patent, and 250+ delivered AI systems.
And last Wednesday, I spent 15 minutes googling how Python dictionary comprehensions work. Because I forgot. Imposter syndrome in AI is brutal because the field genuinely moves faster…
By Pranay Mahendrakar Read postAI regulation isn't the enemy of innovation.
The EU AI Act is now enforced. India's Digital India Act is taking shape. US executive orders on AI are expanding. And sector-specific regulations in healthcare and finance are…
By Pranay Mahendrakar Read postHere are my predictions for AI by 2030. I'll be happy to be wrong on any of them.
AI agents will handle roughly half of routine knowledge work — scheduling, basic research, data entry, report generation. Not because they're perfect, but because they're good enough and…
By Pranay Mahendrakar Read postI learn new AI concepts in about 7 hours.
Hour 1: READ — a paper or detailed blog post for the theory. Not a 10-minute summary. The actual theory. Understand WHY it works, not just THAT it works. Hour 2: WATCH — a tutorial or…
By Pranay Mahendrakar Read postI built a programming language from scratch. It's called Phoenix.
Why? Not because the world needed another language. But because building one teaches you things nothing else can. When you build a lexer, you understand how computers read code. When you…
By Pranay Mahendrakar Read postYou've read through all of this. Now comes the part that actually matters: what are you going to DO?
Not next month. This week. Pick ONE skill that excited you from what you've read. Not three — one. Start ONE project around it. Follow 10 people in the AI community who are building in…
By Pranay Mahendrakar Read postThe biggest misconception about AI is that it's here to replace people.
After building 250+ AI systems across industries — defense, healthcare, manufacturing, finance — I can tell you with certainty: the best outcomes happen when AI handles the repetitive…
By Pranay Mahendrakar Read postThe best ML advice I ever received, from a senior engineer on my first project:
"Stop watching tutorials. Build something that breaks. Fix it. Build something harder. Repeat." I had been stuck in tutorial hell for months — watching course after course, feeling…
By Pranay Mahendrakar Read post