At Practo, I developed, tested, and maintained Python/Django backend solutions. I debugged a production query plan issue on a Java/JPA endpoint, cutting p50 latency from 2.4 seconds to about 1.5 milliseconds.
I also fixed a production race condition that caused duplicate records under concurrent load, using row-level locking and validating the change with an eight-thread regression test. For a MySQL timeout, I traced the issue to an implicit type mismatch disabling an index and resolved it with an explicit cast and EXISTS semi-join.
During the S.N. Bose Internship Programme, I built a Rasa-based conversational AI agent for MEDYOU Health Platform that handled over 100 health-related queries. I integrated intent classification and entity extraction with backend REST APIs, and implemented OTP-based authentication from requirements through deployment.
In my PrivateGPT project, I built a privacy-focused application that anonymizes sensitive information before prompts are sent to external AI models. I developed a two-stage LLM pipeline using Vicuna (LLaMA) for local PII redaction, with FastAPI, Streamlit, and SQLite.
