At Codet.ai, I lead research for an AI platform and developed and extended CHHG, a self-reasoning neurosymbolic knowledge graph on YugabyteDB. I also built an ontology-aware RAG pipeline that ingests long documents with dual-agent deliberation and source grounding.
I built Codet's Omni Box, an event-sourced, CQRS layer for conversations, identity, consent, and outreach workflows. I also developed a social-DM agent with Python, FastAPI, and LangGraph, and designed the integration of the Nina outbound voice agent with the Omni database.
Earlier, at Axillio, I fine-tuned and evaluated speech, video, and vision models, improving model accuracy from 60% to 93% and reducing video generation time from two days to ten minutes. My water-quality AI pipeline using vision-language models and LLMs placed second at TeknoFest 2024, and my research on monitoring contaminated water was accepted at IEEE INMIC 2024.
