I'm a final-year Computer Science student at Dayananda Sagar University, Bengaluru, focused on building full-stack applications backed by AI/ML — not just experimenting with models, but integrating them into production-style systems with real architecture, real data flows, and real tradeoffs.

Each project below was built end-to-end: data layer, backend logic, integration with external APIs/models, and a working deployed product.

• Built Tripzo, a ride-booking platform with a custom ML fare-prediction microservice and full payment integration
• Built InterviewX, a voice-driven AI mock interview platform with dynamic question generation and structured AI evaluation
• Built Neural Vulnerability Detector, a fine-tuned CodeBERT model for source-code vulnerability detection, deployed across three surfaces
• Currently exploring LLM application architecture, RAG pipelines, and agentic systems
• Open to roles in AI/ML Engineering or Full-Stack Development

Full-stack ride-booking app with transparent, ML-driven fare prediction.

• Three independent services: React Native app, Node.js backend, Python ML microservice
• Fare-prediction model using seven real-world inputs — distance, traffic, weather, surge, vehicle type, time
• Full fare breakdown shown to the user rather than a single opaque number
• Razorpay payment integration with cryptographic signature verification
• Four Google Maps APIs in use: Directions, Distance Matrix, Places, Maps SDK

Stack: React Native · TypeScript · Node.js · Python · FastAPI · Scikit-Learn · PostgreSQL · Razorpay

Full-stack, voice-driven mock interview platform that generates personalized technical questions in real time and produces AI-evaluated feedback on candidate performance.

• Dynamic question generation via Google Gemini 2.5 Flash, scaled by role, experience level, and tech stack — no static question bank
• Fully voice-driven interview flow: AI asks questions via TTS, candidate answers by microphone, Deepgram transcribes in real time
• Post-interview evaluation across multiple categories (technical knowledge, problem solving, communication, confidence) with a written summary, strengths, and improvement areas
• Server-validated session authentication via Firebase Admin SDK with HTTP-only cookies; all interview/feedback data scoped per-user at the query level
• Firestore schema with interview and feedback data deliberately separated, linked by interviewId, to keep history lightweight and feedback independently queryable

Stack: Next.js 15 · TypeScript · Tailwind CSS · Firebase (Auth + Firestore) · Google Gemini · Deepgram (STT/TTS)

Fine-tuned CodeBERT model that detects real security vulnerabilities in C/C++ source code, trained on 190,000+ functions from real CVEs.

• Transformer-based semantic vulnerability detection — catches patterns rule-based scanners miss entirely
• F1 score of 0.782 and recall of 0.722 on a held-out test set of 33,000+ unseen functions
• Detects CWE-119 (buffer overflow), CWE-416 (use-after-free), CWE-476 (null pointer dereference), and others
• Deployed across three surfaces: a FastAPI inference server (~250ms GPU inference), a GitHub Action that auto-scans pull requests with inline CWE comments, and a VS Code extension with real-time in-editor warnings
• Exported to ONNX for platform-independent deployment

Stack: Python · PyTorch · Hugging Face Transformers · FastAPI · ONNX · TypeScript · GitHub Actions

• Email: tharunchandru88@gmail.com
• GitHub: @Tharunnxx