LLM-Powered Retrieval Assistant (WIP)
Heading-aware chunking RAG with vector DB backends (Milvus/Qdrant) and strict evaluation to minimize hallucinations.
I design and deliver end-to-end AI solutions across insurance, e-commerce, and healthcare—emphasizing robust APIs, rigorous evaluation, and production-ready deployment.
Heading-aware chunking RAG with vector DB backends (Milvus/Qdrant) and strict evaluation to minimize hallucinations.
Developed LangChain-based multi-agent chatbots for fashion & retail brands, enabling product discovery, order tracking, and returns automation.
End‑to‑end fraud scoring with model monitoring and alerting. Improved detection accuracy and reduced manual review.
GBM‑based churn model with explainability and actionable triggers for retention workflows.
Unsupervised anomaly detection on anonymized CT scans; latent‑space visualization with UMAP/t‑SNE.
Trained on limited data; achieved strong early‑stage classification performance using GANs.