Ayush Yadav

- Built a real-time AI governance proxy that intercepts and validates all requests between applications and LLM APIs (Gemini) before they reach the model.

- Engineered a PII detection and redaction engine using Microsoft Presidio capable of identifying and masking 50+ sensitive data types including emails, credit card numbers and API keys.

- Implemented a semantic prompt injection detection system to block jailbreak attempts and system prompt overrides in real time with sub-50ms latency overhead.

- Built a community-driven parking finder for Kolkata using Next.js, Supabase, and Prisma, featuring an interactive Leaflet/OpenStreetMap interface where users browse and filter live spots by vehicle type, price, and verification status.

- Integrated the OpenStreetMap Nominatim API for geocoding and reverse geocoding, with city-bounded search (viewbox biasing),in-memory caching to reduce redundant requests, and graceful fallback when no match is found — enabling free-text location search without a paid API key.

- Designed a no-auth crowdsourcing submission flow where users drop a pin, search by address, and upload photos to add new spots directly to a Supabase-backed database, with all community submissions flagged " Unverified " pending review.

- Developed an automated digital forensic tool using Selenium WebDriver to simulate human interaction for high-fidelity data extraction from social mediaplatforms.

- Architected a modular desktop application with an Electron.js frontend and Node.js backend, achieving a 90% reduction in mean processing time compared to manual investigation.

- Engineered an automated reporting module using PDFKit to generate comprehensive forensic reports withembedded screenshots and verified timestamps.