Pranjal Singh
Experience
AI Engineer Intern
Stealthera Innovations Pvt. Ltd.
Remote
Building and evaluating speech recognition and voice intelligence pipelines using Whisper and related ASR frameworks for real-time AI-powered voice solutions.
Collecting, cleaning, and preprocessing voice and audio datasets; detecting speech patterns, anomalies, and behavioral indicators from raw audio data.
Fine-tuning AI models and optimizing inference performance; documenting experiments, model benchmarks, and technical findings for weekly review sessions.
Open Source Contributor
AOSSIE (Australian Open Source Software Innovation & Education)
Remote
Contributed to community-driven open-source repositories under AOSSIE, resolving reported bugs and implementing feature requests in full-stack web applications.
Collaborated with global maintainers via GitHub, participating in code reviews, issue triaging, and pull request discussions to improve codebase quality.
Refactored existing modules to improve readability and performance while adhering to project-specific contribution guidelines.
Education
Indian Institute of Information Technology (IIIT), Ranchi
B.Tech
Jharkhand, India
Skills
Projects
Cognitive Voice Intelligence Platform
Next.js, FastAPI, PostgreSQL, Faster-Whisper, Docker
Engineered a clinician assessment dashboard (Next.js, TypeScript) connected to a FastAPI backend, enabling cognitive risk evaluation from speech and acoustic biomarkers.
Optimized an asynchronous transcription pipeline using Faster-Whisper with CTranslate2 for fast inference, processing raw audio through FFmpeg and serving results via OpenAPI/Swagger-documented endpoints.
Designed a PostgreSQL schema with SQLAlchemy and asyncpg for non-blocking storage of session-level linguistic metrics; deployed the frontend on Vercel and the ML inference service on Hugging Face Spaces via Docker.
AI Agent Management Platform
Next.js, TypeScript, tRPC, TanStack Query, Drizzle ORM, Neon DB
Architected end-to-end type safety from database to UI using tRPC with inferRouterOutputs, eliminating all manual type duplication across the full stack.
Secured all CRUD endpoints by engineering custom protectedProcedure tRPC middleware that validates user sessions and cookies before any database operation.
Eliminated stale-data issues by integrating TanStack Query’s targeted invalidateQueries cache invalidation, delivering real-time UI updates without full page reloads.
CarePulse AI – Healthcare SaaS
Next.js, Node.js, Supabase, Vapi SDK, Twilio, Gemini API
Built an outbound AI voice-calling pipeline integrating Vapi and Twilio webhooks to dispatch dynamic, context-aware Hinglish/English calls for medication adherence tracking.
Designed a persistent Supabase PostgreSQL state manager syncing a multi-route dashboard with active cron schedules (is active toggles) and daily compliance matrices.
Integrated a post-call Gemini API audit ledger to parse transcripts into structured compliance reports, achieving 100% TypeScript safety across the mono-repo.
Thread AI – Chrome Extension
TypeScript, Chrome MV3, Gemini API, Vite, CRXJS
Engineered a Manifest V3 sidebar injected into any active tab, enabling zero-friction sub-chat threads from highlighted text/code without breaking the user’s workflow.
Implemented a direct SSE streaming pipeline from the content script to the Gemini API endpoint, delivering real-time token responses with no backend proxy.
Managed secure cross-layer state using chrome.storage.local to persist API keys and coordinate UI transitions across background and content scripts.
Achievements
Smart India Hackathon (SIH)
Qualified internal rounds to represent IIIT Ranchi at the national level.
Problem Solving
Solved 200+ DSA problems on LeetCode and CodeChef; active in global CodeForces contests.
Research
Experimenting with open-source LLM implementations; focus on LLM architectures and AI Automation.