CS student at Guru Gobind Singh Indraprastha University, India.
I'm currently learning system design.
I build production-focused AI systems and scalable backends. I prefer small, testable components that ship and learn from real users.
python c++ javascript node pytorch fastapi postgres docker git linux
SachAI - AI Hallucination Detection · python, FastAPI, HuggingFace
RAG pipeline with hallucination detection middleware.
smallGPT · python, PyTorch
Decoder-only Transformer implemented from scratch.
Multilingual Voicebot · FastAPI, embeddings, serverless
Serverless RAG voicebot cross-language retrieval, real-time TTS, and scalable inference.
Crush Proposal Website · Vercel, AI agent
Personalized proposal site generated with an AI agent and deployed to Vercel. (Success rate: 99.99%)
Every graphic here is generated, not embedded from anyone else's server.
ascii.svg is a photo pushed through a character ramp by
scripts/make_portrait.py; the stat graphics and
these section headings are drawn by a scheduled action
straight from the GitHub GraphQL API, once a day, committing only what changed.
They animate with SMIL inside the SVG, because GitHub strips scripts from
READMEs — and since nothing loads from a third party, nothing here can
rate-limit or go dark. The headings are SVGs for the same reason: GitHub also
strips CSS, so an image is the only way to put this page's own typeface on them.
The typeface is JetBrains Mono, subset to just the characters
each graphic draws and inlined as base64. That isn't only for looks: the
portrait's grid assumes an advance width of exactly 0.600 em, and a viewer whose
default monospace is narrower would otherwise see it squeezed.
Language totals cover public repositories only. year.svg uses the portrait's
character ramp: : + # @, quiet to loud.