i build retrieval engines, backend infrastructure, and AI tools under tight hardware and latency constraints.

most of my work is about making models and systems useful when memory, money, or compute run out. i care more about TTFT, context window management, zero-downtime persistence, and clean abstractions than benchmark scores.

i default to lower-case, first-principles code, and software that doesn't break when real users touch it.

bits&bytes™ (GOBITSNBYTES FOUNDATION) — Chief Technology Officer & Tech Lead. Built the technical stack from scratch; running hackathons, dev squads, and product launches for student builders nationwide.

Previously Research Engineer at jhana (Nov 2025 – Jan 2026): context-window manager for long legal dictations, multimodal consultation analysis, and real-time WhatsApp case intake handling 1.6M+ quarterly requests.

Languages: Python, TypeScript, SQL, Bash, Dart
Backend & APIs: FastAPI, Next.js (App Router), Node.js, Express, REST & WebSockets, Pydantic v2
Retrieval & LLMOps: FAISS, NetworkX, BM25, TEI, vLLM, Cross-Encoders, Vercel AI SDK, Docker
Databases & Storage: PostgreSQL (Neon), SQLite (WAL), Redis, Drizzle ORM, Prisma
Cloud & Operations: AWS (EC2/S3), RunPod, Vercel, pnpm, CI/CD, Linux, Git
Hardware & Edge: Raspberry Pi, ESP32, Pygame/NumPy audio pipelines

• tri-signal retrieval fusion (combining vector HNSW, graph topologies, and BM25 lexical scores)
• adaptive time-series prediction engines on noisy user data with cold-start priors
• self-hosted LLMOps, reducing TTFT and inference costs without accuracy loss
• zero-downtime, local-first database persistence (.mind bundles & SQLite WAL)
• student builder networks & un-gatekept tech education