At PICT, I built a diarized speech analysis pipeline using OpenAI Whisper and Pyannote.audio, reducing manual review time from 45–60 minutes to under five minutes per meeting recording. I also engineered a three-key API fallback system that eliminated user-facing errors during Groq API rate limits across three testing cycles.
I extended that internship project into VoiceDNA-IIL, a proposed framework combining speaker embeddings, LLM-driven role extraction, and vocoder artifact detection for voice-based fraud analysis. The publication was accepted to the 3rd Global AI Summit.
In AI for Bharat – GridBrain AI, I built a full-stack grid analytics platform with demand forecasting and geospatial EV station recommendations across 22 Bengaluru zones. Our optimization model used MILP and LLaMA 3 to generate operator insights, and our team secured a top-three position in our theme.
My projects also include a clinical-trial pre-screening agent, an automated video dubbing pipeline, and an ongoing privacy-first AI digital clone. Curiosity drives the projects I build, and I enjoy exploring unfamiliar domains across AI-assisted systems, backend engineering, and full-stack development.
