Projects
Case studies
What I built, how it was put together, what broke, and what I'd change.
Speech Emotion Recognition
CNN + LSTM model trained across four datasets — 93.27% held-out accuracy, under 200 ms on CPU.
Read the case studyCCTV Analytics for Crime & Crowd Management
Smart India Hackathon 2023 finalist — real-time YOLO detection and deduplicated alerts across four feeds.
Read the case study
Other public projects
- Agentic Honey-Pot — AI Counter-Scam Platform
Built at GUVI AI Impact Buildathon 2026. LLM-driven personas engage live scammers in real time to stall them and extract threat intelligence. FastAPI orchestration with persona/strategy switching and semantic scam-probability scoring; React 19 + Vite dashboard with API-key auth and rate limiting.
Want something like this built?
See the services these projects map to, or get in touch directly.