Yogaselvan S
AI Developer building RAG systems, LLM integrations.
Ready for work
Yogaselvan is ready for their next project!
Retrieval system over 370 sections of Indian tax law, chunked to section level so every answer carries a citation. Evaluated on 52 hand-verified questions baseline recall@5 of 0.71. Diagnosed the failure mode as a vocabulary gap between everyday phrasing and legal terminology, fine-tuned the embedding model on 1,361 pairs, and improved recall@5 to 0.85. Deployed torch-free (ONNX Runtime) to run within free-tier hosting limits.
1
27
Built a real-time hand-landmark detection system using MediaPipe Hands and OpenCV, tracking 21 keypoints per frame from live webcam input. Optimized frame processing for real-time computer vision performance.
0
8
Built an end-to-end pipeline that discovers local businesses without a website, enriches their contact details, and drafts personalized outreach using an LLM. Discovery is idempotent, keyed on a stable source ID, so no business is ever contacted twice across repeated runs. Includes a human-approval gate before any message is sent and output sanitization to stop model-generated commentary from reaching recipients. Runs unattended on a schedule, standard library only, and no ongoing infrastructure cost.
1
22
Built a machine learning pipeline for early Alzheimer's disease detection from structured clinical data, including preprocessing and feature engineering. Trained and evaluated classification models using accuracy-based metrics.
0
4