⚡ AI Hardware Acceleration
Local RAG PDF Search on AMD iGPU (ROCm)
Built a privacy-first, zero-cloud semantic search engine that ingests large multi-page PDF documents, extracts embeddings locally, and queries vector indices in sub-200ms query latency.
- Achieved 94% Top-k retrieval precision over a 10,000+ vector chunk database.
- Reduced query latency by 45% using FAISS HNSW indexing and customized memory pool allocation.
- Zero external API dependencies — 100% data privacy and zero cloud cost.
📊 Verified Metrics
Retrieval Accuracy: 94% Top-k
Query Latency: < 200ms
Speedup vs CPU: 3.2x GPU
Document Scale: 10,000+ chunks
🔒 Security & Identity
Production User Authentication API Service
Architected an enterprise-grade auth backend supporting dual-token (JWT access + refresh) rotation, bcrypt password hashing, and Redis-backed session revoking.
- Built full asynchronous database layer with PostgreSQL & SQLAlchemy async sessions.
- Integrated Redis blacklisting for instant token revocation on logout.
- Delivered 42+ unit & integration test suite achieving 100% endpoint coverage with Pytest.
📊 Verified Metrics
Pytest Coverage: 42+ Test Suite
Session Revoke: < 2ms (Redis)
Password Hashing: Bcrypt (Cost 12)
DB Migration: Alembic Sync