AI engineer, final year CSE UG student — i build the system around the model. retrieval, agents, orchestration, evals.
a few builds so far: a Graph-RAG project manager, a collaborative editor with a sync engine i wrote myself, and a scoring model i caught overclaiming and rewrote around that. they're in the repos below.
🛠 working on
- long-horizon agents that don't lose the thread
- memory and tool routing that hold up over a long session
- evals that catch a regression before it ships
🔭 currently obsessing over
- Graph-RAG, and where the graph beats plain vector search
- agent evals — the part everyone skips
- model validation and interpretability
- not letting LLMs do math
