Sahil Deore

I build practical AI products, including Nexus AI Interviewer, a voice-first interview agent powered by Gemini 2.5 Flash. I optimized its STT-to-Gemini-to-TTS pipeline to sub-two-second responses and built automated transcript scorecards.

For TenantWatch, I integrated Hugging Face zero-shot classification into a FastAPI backend to route tenant complaints across four categories with 90%+ accuracy. I also designed confidence-based manual triage and an interactive geocoded complaint heatmap.

I fine-tuned a MobileNetV3 Large model for LeafShield AI, reaching 95.6% validation accuracy across 38 crop-condition classes. Its FastAPI inference server returns predictions and treatment recommendations in under 100ms on CPU.

Alongside AI work, I deliver full-stack web applications for commercial clients and have independently launched production websites for Dosa Plaza subsidiaries. I've also led developer teams, scoped APIs, reviewed code, and supported live platforms handling real transactions.