AI + backend systems. I started out training models, then kept having to build the API, the database, the auth, the deployment around them — and that's the part I ended up caring about most.
Fourth-year B.Tech in AI & Data Science at Shiv Nadar University Chennai, also doing a BS in Programming and Data Science at IIT Madras. Research intern at the IIT Madras Computational Imaging Lab, working on lensless camera systems and physics-informed deep learning.
Portfolio: ashvp.com
🛡️ GitGuard — an AI-powered Git CLI that assesses risk before you run something destructive
Natural-language Git commands, AI-assisted commit messages, risk assessment before execution, checkpoints and rollback for when a command goes wrong. Published as a Python package — 2.9k downloads.
Python Git CLI LLM APIs
📚 AI Question Paper Generator — document-grounded question generation, not an LLM guessing
PDF → chunk → embed → retrieve → generate → validate. The retrieval step is the point: questions come from the source material instead of the model improvising.
FastAPI LangChain FAISS Embeddings
👁️ Mobile Attendance System — a face-recognition experiment that became a full platform
Auth, verification, groups, attendance workflows, admin panel — mobile client backed by a real API and persistent database state.
FastAPI PostgreSQL SQLAlchemy Computer Vision Docker
📄 AI Resume / Document Intelligence — making structured documents machine-readable and editable
Combines document extraction, external data sources, and LLM reasoning into document-generation workflows that hold up to actual editing, not just one-shot output.
LLMs Document Processing Structured Generation
🦠 Android Malware Analysis — the other half of building software: understanding how it gets abused
Static analysis of Android applications — Python tooling on top of Androguard, backend APIs for orchestration.
Python Androguard FastAPI ADB
🤖 Personal AI — treating a personal assistant like an actual system, not a chat window
Ongoing project on tools, memory, context and automation for an assistant that behaves like real software rather than a wrapper around an API call.
LLMs Agents Memory Automation
Agentic AI, RAG systems, LLM application architecture, and figuring out how to make AI-powered projects boring enough to actually trust in production.
3D graph generated via github-profile-3d-contrib — set up as a scheduled GitHub Action, so it updates itself. Setup notes below.
Setting up the 3D contribution graph (one-time)
Add a workflow file at .github/workflows/profile-3d.yml in your ashvp/ashvp repo using the yoshi389111/github-profile-3d-contrib action on a daily cron schedule, output to an output branch. It replaces the snake — same "auto-generated GitHub art" idea, but isometric instead of a grid crawl.