ML engineer shipping production Python at a prop trading firm. Models, services, and the infrastructure that puts the two together.

I do applied ML at a place where the model's output becomes a real order — which is a very effective way to learn what "good enough" isn't. Most of my work sits at the seam between a model and a production system: the eval harness, the serving layer, the latency budget, the human override. When the unit is dollars, off-by-one errors and silent regressions both get loud fast.

Today that work lives inside Axxela, a prop trading firm. The stack is unfashionable in the best way — Python, Flask, MySQL, Redis, a few well-placed background workers, and a TensorFlow / scikit-learn / Llama-Index ML layer wrapped in services that recover cleanly. I own the loop end-to-end: schema, model, evaluation, deploy, and post-incident review.

Before Axxela I led ML at Athena Education and built internal tooling at BNP Paribas — first taste of code running against real money. Earlier still, two mobile internships (Flutter at HIH7, Java/Android at RevMeUp) where I learned to read a stack trace under deadline. The thread is the same: production code, real users, and the discipline that comes from knowing exactly what happens when something breaks.

I'm currently open to senior ML / ML-platform / applied-ML and backend roles. Kolkata-based; remote, hybrid, or relocation considered.