Backend engineer who wandered into ML and never came back. Started with Java + Spring Boot, now I build RAG pipelines and agentic systems — with enough LLM internals under the hood to know why things break, not just that they did.
I go deep before going wide: wrote backprop in NumPy before touching PyTorch, built BM25 from scratch before importing a library. Not because anyone asked. Because reading the docs felt like cheating.
What I do: RAG pipelines, agentic systems, backend that doesn't fall over — and enough handle on transformer internals to reason about the model, not just call its API.
Stack: Python · Java · FastAPI · Spring Boot · LangGraph · Docker · Postgres
Projects, demos, and the receipts: → sonuverma.vercel.app
📫 tannsonuverma@gmail.com · LinkedIn
"It worked on my machine" is not a personality, but it is my origin story.
