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.
352 posts
I bombed my first ML interview. Completely.
They asked me to design a recommendation system at scale and I rambled about model architectures for 20 minutes without once mentioning data pipelines, serving infrastructure, or…
By Pranay Mahendrakar Read postIf you're still serving ML models with Flask in 2026, we need to talk.
I migrated a client's model serving from Flask to FastAPI last quarter. Same model, same hardware. Results: 3x more concurrent requests handled, auto-generated API docs that the frontend…
By Pranay Mahendrakar Read postPeople keep asking me: "Is AI a bubble?"
Honest answer: parts of it, probably yes. But AI itself? No. The parallels to the dot-com era are real. Insane valuations for companies with no revenue. "AI-powered" slapped on products…
By Pranay Mahendrakar Read postWhen I started in NLP, the exciting project was sentiment analysis — figuring out if a review…
That's a homework assignment now. NLP in 2026 is building conversational AI that remembers your previous conversations. It's systems that read a 200-page legal contract and flag…
By Pranay Mahendrakar Read post"But it works on my machine!"
If I had a rupee for every time I heard this during model deployment, I'd retire. Docker solved this problem for software engineering a decade ago. Yet I still see ML engineers in 2026…
By Pranay Mahendrakar Read postThe CTO of a mid-size company told me something interesting last month:
"I don't care about your model's F1 score. Tell me how it saves us money." That conversation changed how I pitch AI projects. Engineers love talking about architectures and benchmarks…
By Pranay Mahendrakar Read postTime series is the most underrated specialization in ML and I will die on this hill.
Think about how much of the world is sequential data: stock prices, server metrics, patient vitals, weather patterns, energy consumption, IoT sensor readings, factory equipment…
By Pranay Mahendrakar Read postLet's talk about AI salaries honestly, because there's a lot of misinformation out there.
In India right now — entry level (0-2 years) for AI/ML roles: 8-15 LPA. Mid level (2-5 years): 15-35 LPA. Senior (5-8 years): 35-60 LPA. Lead/Principal: 60 LPA to 1 Cr+. Globally, US ML…
By Pranay Mahendrakar Read postHere's something most people don't realize: every time you use ChatGPT, you're using…
RLHF — Reinforcement Learning from Human Feedback — is what makes LLMs actually helpful instead of just statistically completing text. It's RL that transforms a base model into something…
By Pranay Mahendrakar Read postI review AI portfolios when we hire. Here's what makes me instantly interested vs. instantly bored.
Instantly bored: Titanic survival prediction. MNIST digit classifier. Any project where the first Google result is a step-by-step tutorial. Instantly interested: something original that…
By Pranay Mahendrakar Read postWhat if you could train an AI model across 50 hospitals without any patient data ever leaving…
That's federated learning. And it's one of the most underappreciated technologies in AI. The idea is elegant: instead of bringing data to the model, you bring the model to the data. Each…
By Pranay Mahendrakar Read postA junior engineer on my team asked me nervously: "Will AutoML make my job obsolete?"
I laughed. Not at them — at the question, because I asked the same thing five years ago. Here's what actually happened: AutoML made me 10x faster, not 10x less valuable. AutoGluon…
By Pranay Mahendrakar Read post