Hi, I'm Indranil, a painter.
I build agents and ML on graphs and language models. I'm an ML engineer at a major US telecom, where dynamic agents handle multi-modal inbound/outbound campaigns to fault resolution end to end. At UT Dallas I research adversarial red-teaming agents for network protocols. Before that, graph ML and recommenders at G2and transformers in production at Byju's. Outside of work, I also...
- Write about agents, graph ML, and what I'm learning.
- Prototype local-first tools: a personal inference engine, protocol red-teaming agents, and a notes-to-knowledge-base workflow in Obsidian.
- Contribute to OpenClimateFix's open-source graph weather models.
- Paint, and read the Stoics when things get loud.
Latest
- The Judge Said Yes, Then Explained Why Not
- Same Bot, Two Games
- The Harness Is the Product
- Two Stores, One Truth
- Four Results I Had to Retract
- My Stateful Memory Made the Fuzzer Worse
- What Two GPUs at Home Taught Me
My Favourites
7 posts, 4 prototypes, 11,459 words, and countless hours.
Selected Prototypes
- 5GAgent: An LLM-driven tester for 5G protocols: it reads 3GPP specifications, builds a state-machine graph, proposes adversarial message sequences, and runs them against a real testbed.
- Personal Inference Engine: A home GPU server with two consumer GPUs (an 8 GB card and a 24 GB card) that serves local models to my own automation projects, reachable from anywhere over Tailscale.
- OpenClimateFix Weather Pipeline: Solar forecasts decide how much gas a grid operator burns as backup.
- Three.js Experiments (2018): Before the graphs and the language models there was this: a pile of little Three.js experiments I built for an internship task at IIT Bombay, back in 2018.