I study AI and ML at VIT in Bhopal (Integrated M.Tech, final year) and write software for SAITS, a Swiss AI services company. Most of what I build puts a model inside something people use: agents that call real tools, photo search that stays on your own machine, a vault that hands a secret to a script only while it runs.

Right now I'm fine-tuning small open LLMs.

Portfolio · Writing · Résumé · LinkedIn · X

• PentaVault: runtime secrets for AI-assisted development. Secrets stay encrypted, access is deny-by-default, and the CLI injects a value only while a command runs. (repo)
• CareerOS: a career counsellor students can talk to. A voice agent interviews them, books sessions on their Google Calendar and answers only from official sources. (live · repo)
• Find: search your photos in plain language, fully self-hosted, so nothing leaves your machine. (repo)
• NexusFlow: a canvas workflow designer that flags a broken process while you draw it, then simulates it step by step. (live · repo)
• Allo: timed inventory holds where two carts can never claim the same last unit, enforced with Postgres row locks. (live · repo)

More, including smaller builds, on my projects page.

• Dokploy: I maintain my own build of this self-hosted PaaS and run my servers on it. 138 commits and 16 releases add fleet SSH, a firewall that rolls itself back if it locks me out, an encrypted vault and restic backups with restore drills.
• mem0: a self-hosted memory server with an entity graph, webhooks and a one-container pgvector image, synced with upstream every week.
• recodehive: 7 merged PRs, including Clerk sign-in, Algolia search and a fix that stopped exposing tokens in the browser.
• Gityzer: 9 merged PRs during Social Winter of Code 2025, among them the AI badge generator and SVG rendering. Second-largest contributor.
• Nyay Sathi: my v2.0 upgrade moved this Indian-law assistant to agentic RAG with cited sources.

Python, TypeScript and Rust. PyTorch, Hugging Face and LangChain for models; Next.js, FastAPI and Fastify for apps; PostgreSQL with pgvector, Redis and Convex for data; Docker, Linux and GitHub Actions to ship it.