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
The model that matters isn't Gemini 3.7 Flash. It's the 3B one on your phone.
I want to talk about the model nobody announced with a keynote: the 3B parameter model sitting on your phone right now that outperforms the 7B cloud model your team was calling…
By Pranay Mahendrakar Read postI spent last weekend building an AI agent that books my flights, compares hotel prices, and…
Two years ago, that would've taken a team of three and two months. Here's what nobody's talking about with agentic AI — it's not the technology that's hard. LangGraph, CrewAI, AutoGen…
By Pranay Mahendrakar Read postI interviewed 23 candidates for an ML engineer role last quarter.
19 of them had almost identical resumes — TensorFlow certification, Coursera ML specialization, a Titanic dataset project on GitHub. The 4 who stood out? They'd actually built something…
By Pranay Mahendrakar Read postSomething clicked for me last month when I was debugging a multimodal pipeline.
I fed the system a photo of a whiteboard from a meeting, an audio recording of the discussion, and the follow-up email thread. It synthesized all three into a coherent summary that was…
By Pranay Mahendrakar Read postA factory I consulted for was sending 2TB of camera footage to the cloud daily just to detect…
Their cloud bill was astronomical. And there was a 3-second delay between a defect appearing and the alert firing. In manufacturing, 3 seconds means 6 defective units already packaged…
By Pranay Mahendrakar Read postHot take: 95% of "LLM applications" I see on LinkedIn are demos, not products.
The demo: "Look, I connected GPT to my database and it answers questions!" The production system: It handles 10K concurrent users. It doesn't hallucinate on financial data. It costs…
By Pranay Mahendrakar Read postThe most depressing statistic in ML: 80% of models never make it to production.
I've seen this firsthand. A brilliant data scientist spends 3 months building a model with 94% accuracy. Everyone's excited. Then it sits in a notebook for 6 months because nobody knows…
By Pranay Mahendrakar Read postA friend in cybersecurity told me something that kept me up at night.
"The attackers are using AI now. They're generating thousands of unique phishing emails per hour, each personalized to the target. No typos. Perfect grammar. Contextually relevant." This…
By Pranay Mahendrakar Read postI don't have a PhD. A lot of people have opinions about that.
When I was starting out, I was told — repeatedly — that you can't be a "real" ML engineer without a doctorate. That I'd hit a ceiling. That I'd never be taken seriously. 250+ AI systems…
By Pranay Mahendrakar Read postI built a RAG system last year that worked perfectly on my test data.
Turned out my chunking strategy was splitting financial tables right down the middle. Half the revenue data ended up in one chunk, half in another. The retrieval grabbed one half, and…
By Pranay Mahendrakar Read postThe wildlife detection system I built for railway tracks at IIRS-ISRO taught me something that…
In the lab, my model detected animals at 96% accuracy. On actual railway footage? It dropped to 71%. Rain, fog, nighttime, motion blur — real-world conditions are brutal. I spent three…
By Pranay Mahendrakar Read postConfession: I used to write terrible Python.
Jupyter notebooks with cells numbered out of order. No type hints. Global variables everywhere. Functions called "process_data_v2_final_FINAL." Sound familiar? The turning point was when…
By Pranay Mahendrakar Read post