Sole MLE on a 10-person pre-launch team (founder + 8 SWEs). Transitioned from senior-year intern to full-time post-graduation, owning all ML architecture, research, and production decisions in direct collaboration with the founder ahead of launch.

• Shipped production attribute and unit harmonization: SBERT embeddings over 125K+ product attributes, cosine-similarity retrieval, GPU-accelerated batch inference, with confidence metadata and top-k match evidence surfaced for output explainability.
• Designed schema-agnostic ingestion that auto-detects multiple JSON formats and normalizes them into a canonical attribute representation; built unit-harmonization logic across physical, electrical, and thermal dimensions, with output now consumed by the company's database, browser extension, mobile app, and web platform.
• Built a real-time ML operations dashboard (FastAPI + WebSockets) with batch upload, step-level event streaming, and grouped downloadable outputs for debugging and stakeholder demos.
• Prototyping a multi-agent harmonization and product-data scraping pipeline built with Claude Agent Skills orchestrated through Obsidian, and presenting architecture deep-dives to the company's advisory board (specifics under NDA).

• Scoped, specified, and deployed an IoT monitoring system (Raspberry Pi 4 with thermocouples and potentiometers) to track pipe temperature and thermal expansion in fossil fuel power plants for predictive maintenance; owned hardware selection, setup, and data visualization end-to-end.
• Researched, specified, and configured NAS hardware and a NAS-to-SharePoint sync workflow to centralize 200GB+ of engineering data across projects, replacing ad-hoc manual transfers.