|
IN THE MATTER OF: ARYAN GUPTA B.Tech ECE, BIT Mesra ('27) · Founder, Madhav AI Jurisprudence Pvt. Ltd. · Ranchi, India |
I build systems that turn messy, high-stakes information into something a professional can actually reason with.
⚖️ Madhav.ai — Justice Reimagined An AI-native legal intelligence and workflow platform for Indian lawyers and law firms — not a chatbot wrapper, but an operating layer that takes a lawyer from raw legal information → understanding → analysis → research → drafting → execution. It's a full platform, not a single tool:
| Module | What it does |
|---|---|
| 🔍 Search Engine | Purpose-built legal search across Indian judgments, cases, and authorities |
| 📚 Research Engine | Deep legal research over a ~1.5 TB legal corpus using RAG + vector retrieval |
| 🗂️ CaseSpace | Per-case workspace — structures evidence, facts, timelines, contradictions, and gaps from uploaded case documents |
| ✍️ Drafting Engine | AI-assisted drafting (bail applications, petitions, etc.), aware of the IPC → BNS transition, verification-first |
| 📋 Universal Drafting Studio | Structured, schema-driven drafting across 60+ Indian legal document types — no generic chatbot flow |
| 🤝 Collaboration | Teams, matters, role-based access, ethical walls, and auditability for firms |
Built on FastAPI · PostgreSQL + pgvector · HNSW · multi-stage LLM pipelines (small models for triage, larger models with validation for deep reasoning) — because a hallucinated legal authority isn't a minor bug. Currently pre-launch, heading toward a private beta with a focused group of litigation and criminal-law practitioners.
🧠 MemoryOS
My flagship open-source project, built on the side — a local-first, model-agnostic, three-tier AI memory library, published on PyPI as memoryos-local. Built so any LLM application can retain context, preferences, and decisions without shipping user data to someone else's server.
Python Local-First PyPI
⚙️ One-time setup for the metrics card above (click to expand)
The old github-readme-stats.vercel.app cards are gone — that shared public instance is chronically rate-limited (it hits GitHub's own API quota because everyone uses the same demo URL), which is why they showed as broken images. This replaces both cards with one self-generating metrics.svg baked directly into your repo, so nothing depends on a stranger's server being up.
Add this file at .github/workflows/metrics.yml in your aryan-iconic/aryan-iconic profile repo:
name: Generate Metrics
on:
schedule:
- cron: "0 */6 * * *"
workflow_dispatch:
push:
branches: [ main ]
permissions:
contents: write
jobs:
metrics:
runs-on: ubuntu-latest
steps:
- uses: lowlighter/metrics@latest
with:
filename: metrics.svg
token: ${{ secrets.GITHUB_TOKEN }}
theme: dark
base: header, activity, community, repositories
config_timezone: Asia/Kolkata
plugin_languages: yes
plugin_languages_analysis_timeout: 15
plugin_languages_limit: 8It commits metrics.svg straight to your main branch on a schedule — no separate output branch needed, and no shared rate limit to hit.
One-time setup for the snake graph below it:
name: Generate Snake
on:
schedule:
- cron: "0 */6 * * *"
workflow_dispatch:
push:
branches: [ main ]
permissions:
contents: write
jobs:
generate:
runs-on: ubuntu-latest
steps:
- uses: Platane/snk@v3
with:
github_user_name: aryan-iconic
outputs: |
dist/github-contribution-grid-snake.svg
dist/github-contribution-grid-snake-dark.svg?palette=github-dark
- uses: crazy-max/ghaction-github-pages@v4
with:
target_branch: output
build_dir: dist
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}It runs itself after that — nothing to maintain.
|
I'd rather ship one real product than demo ten toy ones. Currently building Open to remote roles and open-source collaboration — backend/AI engineering, agents, memory systems, or legal-tech. If that's you, my inbox is open. |