Hey, I'm Aditya. I build things at the intersection of AI and systems. Taking ideas from zero to one, letting curiosity lead the way, and always shipping something new.
PROJECTS
Open-source AI agent skills platform with bidirectional MCP server architecture and Shadows registry.
Wearable AI pendant with persistent contextual memory, connecting via BLE to iPhone for on-device processing.
Mac voice agent that parses intent via LLM and executes UI actions using macOS Accessibility APIs. Top 5 at TreeHacks Stanford.
Contactless vitals monitor from chest-wall micro-displacement using FFT. 95%+ Pearson correlation on Raspberry Pi 4. Top 10 at CalHacks.
Real-time collaborative AI presentation tool with conflict-free sync via OT. Cut per-slide gen from 6.3s to 2.5s. 3rd place at ClaudeHacks.
EXPERIENCE
- — Built an episodic RAG memory pipeline for a nano glasses wearable using FAISS vector indexing and sliding window chunking, with model serving via Vertex AI. Hit sub-50ms on-device query latency with persistent context across sessions.
- — Built the signal processing pipeline for MEDUSA, an NSF-funded 77 GHz MIMO mmWave radar. Used FFT and ICA-based spectral decomposition to recover cardiac and respiratory waveforms, validated at sub-1 BPM accuracy against clinical-grade monitors.