Skip to content
View muskan-khushi's full-sized avatar

Block or report muskan-khushi

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
muskan-khushi/README.md

Muskan Khushi

hey you!

LinkedIn LeetCode Codeforces CodeChef


I have always been more interested in the shape of a problem than the size of a solution.

I sketch before I sleep and solve problems before I eat. I grew up reading stories and somewhere along the way started writing code — turns out they ask for the same thing: precision in the service of something that feels true.

I study mathematics and computing at NIT Patna, third year into a five-year dual degree. What keeps me here is the same thing that keeps me reading at 2am — the unreasonable possibility that the next page changes everything.


✦ things I have built

ARIAagentic regulatory intelligence architecture

OpenEnv Hackathon Finalist · Meta × HuggingFace

I wanted to know if an agent could learn to think like an auditor — not just match patterns, but reason under pressure, hold contradictions, and act anyway. ARIA is the answer I built: the first RL environment for multi-framework compliance auditing, where agents navigate GDPR, HIPAA, CCPA, and SOC 2 across five tiers of difficulty, and at step 25 — mid-audit — a live breach fires. contain it. document it. notify. don't stop auditing. The reward engine has 18 triggers and knows when you're cheating.

Python FastAPI Pydantic v2 React 19 TypeScript WebSockets Docker HuggingFace Spaces


Mule-Hunterreal-time financial fraud detection

Fraud doesn't announce itself. It hides in the texture of ordinary transactions, in the unremarkable node at the centre of a ring that shouldn't exist. I trained a GraphSAGE model on 590K+ nodes to find exactly that — the structures that look like noise until they don't. 94% precision · 89% recall · sub-500ms latency · 67% fewer false positives.

PyTorch Geometric FastAPI Docker NetworkX Pandas


Health AtlasAI-powered provider credentialing

Healthcare credentialing is slow, manual, and consequential. I built a LangGraph orchestration engine that parallelises what used to be sequential — 6 validation nodes, 4.3× faster, a confidence model that routes 85% of cases automatically and knows enough to flag the other 15% for a human who can actually judge. Full provenance. HIPAA-compliant. No shortcuts on the things that matter.

LangGraph FastAPI PostgreSQL Groq Vision API React Tailwind CSS


✦ stack

languages      python  ·  c++  ·  typescript  ·  java  ·  sql
ai / ml        pytorch  ·  pytorch geometric  ·  langchain  ·  langgraph
               gnn  ·  rag  ·  generative ai  ·  agentic workflows  ·  openenv
backend        fastapi  ·  node.js  ·  websockets  ·  docker  ·  kubernetes
data           postgresql  ·  mongodb  ·  chromadb  ·  minio
platforms      aws  ·  huggingface spaces  ·  vercel  ·  github actions

400+ problems across LeetCode · Codeforces · CodeChef — not because I had to, but because I like the quiet of a problem that hasn't been solved yet


✦ recognition

2026   OpenEnv Hackathon Finalist        Meta × HuggingFace
2026   ISEA Cybersecurity Challenge      National Finalist · Top 20 across India
2026   EY Techathon 6.0 Semifinalist     Top 61 of 50,000+ participants
2024   National Hackathon Winner         Best all-girls team · Hackofest, NIT Patna

✦ github


when I am not doing this — music, sketching, chess, books that rearrange something in me

open to conversations about things worth thinking about — graph ml · memory systems · agentic ai · or honestly, anything that keeps you up at night

Pinned Loading

  1. Health_Atlas Health_Atlas Public

    Forked from Rupali2507/Health_Atlas

    Health Atlas— a multi-agent AI platform that autonomously validates, enriches, and prioritizes healthcare provider data. Built with FastAPI, React, and LangGraph, it delivers real-time, VLM-ready a…

    Python

  2. Synapse Synapse Public

    Transform static PDFs into interactive, AI-powered conversational companions. Built with TypeScript, Python, FastAPI, LangChain, ChromaDB, and local LLMs (Phi-3/LLaMA), Synapse leverages Retrieval-…

    TypeScript 1

  3. Rupali-2507/MULE_HUNTER Rupali-2507/MULE_HUNTER Public

    Real-time Mule Account Detection Engine for UPI. Uses Graph Neural Networks (GraphSAGE), JA3 Fingerprinting, and Behavioral Forensics (EIF) to identify money laundering rings, smurfing, and collusi…

    TypeScript 7 4

  4. VeritasStream VeritasStream Public

    AI-powered forensic intelligence platform that detects cyber threats in real-time using hybrid ML (Isolation Forest + pattern matching). Features voice-driven security briefings, immutable evidence…

    JavaScript