Nirman

AI-powered idea-to-execution platform that challenges assumptions, validates with evidence, and guides execution.

Last updated 2026-11-04

Nirman platform dashboard preview — Nirman platform live at nirman-psi.vercel.app

ROLEFull-Stack Developer

STACKNext.js · Python · FastAPI · LangGraph · PostgreSQL · Gemini API

YEAR2026

STATUSActive

LINKSLive Platform Source Code

02MY ROLE

Full-Stack Developer

03README

Nirman

From idea to something real.

Nirman helps people figure out what is worth building before they waste time building it.

Nirman is an AI-powered evidence-to-execution platform that takes a rough idea, extracts its hidden assumptions and uncertainties, helps test those assumptions with real evidence, synthesizes the results, and recommends the next strategic step.

The Core Product Philosophy

"Don't optimize the plan. Optimize the direction."

text

IDEA  →  UNDERSTAND  →  CHALLENGE  →  VALIDATE  →  EVIDENCE  →  DECISION  →  BUILD  →  LEARN  →  ADAPT

Tech Stack

Frontend: Next.js 16 (App Router), TypeScript (Strict), Tailwind CSS, Lucide Icons, React Hook Form, Zod
Backend: Python 3.11+, FastAPI, Pydantic v2, SQLAlchemy (Async), LangGraph
Database: PostgreSQL / Supabase, pgvector support (with zero-config SQLite for instant local dev)
AI Orchestration: Provider abstraction supporting Google Gemini 2.5 Flash with zero-failure local fallback

Monorepo Structure

text

nirman/
├── apps/
│   ├── web/                     # Next.js 16 Web Application
│   │   ├── app/                 # App Router (/, /dashboard, /projects, /settings, etc.)
│   │   ├── components/          # Reusable design system (Card, Button, Badge, AppShell)
│   │   └── lib/                 # Typed API client & Supabase helpers
│   └── api/                     # FastAPI Backend
│       ├── app/
│       │   ├── api/             # REST Routers
│       │   ├── models/          # SQLAlchemy async domain models
│       │   ├── schemas/         # Pydantic v2 validation models
│       │   ├── ai/              # GeminiProvider & MockProvider
│       │   ├── tools/           # 8 Agent Tools in ToolRegistry
│       │   ├── workflows/       # LangGraph 8-Node StateGraph
│       │   └── core/            # Database engine, config, and seed data
│       └── tests/               # Pytest suite
├── packages/
│   ├── types/                   # Shared TypeScript models
│   └── ui/                      # Shared UI primitives
├── supabase/
│   └── migrations/              # Supabase PostgreSQL DDL with pgvector
├── .env.example                 # Environment variable template
├── ARCHITECTURE.md              # Deep-dive system architecture specification
└── README.md

Getting Started Locally

1. Clone & Setup Workspace

bash

git clone https://github.com/Rashmin47/Nirman.git
cd Nirman

# Copy environment variables
cp .env.example .env

2. Run the Frontend (Next.js)

bash

# Install dependencies
pnpm install

# Start development server
pnpm dev

Open http://localhost:3000 in your browser.

3. Run the Backend (FastAPI + LangGraph)

bash

cd apps/api

# Create and activate Python virtual environment
python3.11 -m venv .venv
source .venv/bin/activate

# Install requirements
pip install -r requirements.txt

# Run the API server
uvicorn app.main:app --reload --port 8000

API Documentation (Swagger UI) is available at http://localhost:8000/docs.

Seeded Realistic Demo Project

On first startup, the database automatically seeds a complete, realistic demo project:

Project: Student Freelance Platform
6 Assumptions (including contradicted discovery assumption and supported credibility assumption)
1 Experiment: Discovery vs. Credibility Validation Interview
35 Evidence Records: Realistic quotes from students and small business clients
Synthesized Themes: Portfolio credibility (45%), Escrow trust (22.5%), Discovery (17.5%), Pricing (15%)
Strategic Decision #01: Direction pivot from marketplace job board to Proof-of-Skill Profile & Escrow Escort

Running Backend Tests

bash

PYTHONPATH=apps/api apps/api/.venv/bin/pytest apps/api/tests/test_api.py -v

All 5 core pipeline tests run and pass out-of-the-box.