I built the Orbital Scene Preprocessor, adapting YOLOv8n for six-band multispectral input and exporting it to INT8 ONNX. Its CPU inference turns each image tile into a compact JSON brief.
For the same project, I created a grounded Gemini analyst using FAISS retrieval, SQLite memory, a rule-based fallback, and an LLM-as-judge check. I wrote its 11-test suite and deployed a Streamlit and Folium dashboard.
On the Multi-Agentic RAG AI Research Scientist project, I was the prompt-planning lead in an eight-member team. I designed prompts for four agents, including a safety agent that checks cited claims against retrieved evidence.
I also built DisasterMesh, a wireless-sensor mesh simulation with Dijkstra routing, flood-routing fallback, and self-healing rerouting. I wrote 24 pytest tests and deployed it on Render; as Chair of the IEEE Computer Society Student Chapter at Mahindra University, I lead a 25-member team.
