Built in production, not in tutorials

Owned retrieval quality on a production RAG assistant, from building the evaluation set to shipping the hybrid retrieval rewrite it justified.

• 01Built the retrieval evaluation harness (nDCG@k, MRR, Recall@k) and wired it into CI with Weights & Biases run tracking.
• 02Shipped hybrid BM25 and dense retrieval with reciprocal rank fusion, plus cross-encoder reranking on the candidate head.
• 03Added structured logging, latency and error-rate alerting so regressions surfaced before users found them.

Co-founded a course-aware AI study platform, grew it past 1,000 students, and owned both the product direction and the retrieval stack behind it.

• 01Designed PRASE, the pedagogical retrieval pipeline that kept generated questions inside the course being studied.
• 02Ran the full product loop: user interviews, roadmap, pricing decisions, and the analytics that informed them.
• 03Built and operated the Next.js and FastAPI stack, including payments, auth and cost controls on inference.

Shipped product features across the stack on a small team, with direct ownership of the work from ticket to deploy.

• 01Delivered full-stack features end to end, including API design, data model changes and the interface on top.
• 02Improved reliability of existing services through better error handling and observability.

Advised on technical direction and built against real business constraints, translating between commercial goals and engineering tradeoffs.

• 01Scoped and delivered technical work tied directly to business outcomes rather than a fixed spec.
• 02Worked across time zones with non-technical stakeholders, which is where most of the product thinking happened.

Built and evaluated computer vision models, including the work that became a published paper.

• 01Trained and evaluated deep learning ensembles on medical imaging data at scale.
• 02Published COVID-19 chest X-ray detection research in IJCRR using Xception with EfficientNet-B7 and ResNet152V2 across 11,000+ images.

Perception and machine learning work on an autonomous agricultural robotics team.

• 01Built vision and machine learning components for autonomous field operation.
• 02Worked inside a multidisciplinary team where mechanical and software constraints meet.