Meta Platforms · Senior Staff Software Engineer
Nov 2018 – Present
Technical Lead, Voice AI
2025–Present- Designed end-to-end AI architecture for an ambient audio recall system on wearable devices: on-device audio capture → cloud transcription → LLM summarization (Llama 3) → FAISS vector retrieval — enabling natural language queries over real-world audio memory.
- Built multi-modal retrieval pipeline querying audio, visual, and conversational memory in parallel with cross-domain intent routing; established offline eval pipelines with systematic error analysis, driving measurable recall quality improvements across releases.
- Made key on-device vs. cloud tradeoff decisions: chose on-device speaker diarization (PyAnnote-based) over cloud API, reducing power consumption by ~30%; coordinated 10+ dependent workstreams through multiple firmware freezes to ship on schedule.
Technical Lead, Autonomous Crash Investigation Agent
2025- Built a production multi-agent system that autonomously diagnoses device crashes, performs root cause analysis across logs/coredumps/source code, and generates code fixes — deployed across 4+ wearable product lines and auto-triaging all new crash tasks.
- Architected modular agent design with specialized sub-agents per crash domain (JVM, native, embedded RTOS), each equipped with domain-specific tools: log analysis, crash dump symbolication, GDB-based coredump inspection, and semantic code search via embeddings.
- Built an evaluation framework using historical crash-fix pairs: automated reproduce-and-compare grading via semantic similarity (sentence-transformers + cosine scoring) enables continuous quality iteration. Reduced median crash resolution time from 13 → 3 days.
Technical Lead & Engineering Manager, Systems Performance
2023–2024- Built and managed a global team of 10 engineers driving performance optimization for AI-powered wrist wearables; led cross-org effort to define and ship performance targets for a consumer AI device.
- Built real-time profiling and monitoring tooling for embedded RTOS systems to measure AI workload performance, enabling data-driven tradeoffs between model size, latency, and power on-device.
Technical Lead & Engineering Manager, Display & Graphics
2019–2023- Built and managed a team of 8 engineers; shipped the display software stack on Quest 2, Quest Pro, Quest 3, and Ray-Ban Meta smart glasses — products collectively reaching tens of millions of users.
- Invented and patented a dynamic display brightness and refresh rate modulation system (US20230360566A1) — an adaptive control technique that generalizes directly to resource-aware on-device AI inference.