At Xlofy.ai, I developed four reusable Python pipelines for data ingestion, model integration, evaluation, and testing. I also built regression and classification workflows and standardized more than ten evaluation metrics.
I built VentureLens, a startup research platform that combines Gemini embeddings with cosine-similarity retrieval over source documents. I connected its LLM outputs to FastAPI using structured Pydantic responses and source, page, and chunk citations.
For BlackBox Audit, I developed a voice-agent reliability harness that checked spoken confirmations against backend booking state. Across 606 interruption trials, the audio-fencing state machine reduced mismatches from 150 to zero; I also added 229 automated offline tests.
