Sangam

Grant intelligence for Indian civil-society organisations and the funders who back them.

Live app · Docs · Contributing · Report a bug

Why Sangam

Finding the right funder is slow, manual research: who gives in this state, to this cause, at this size, and have they funded anyone like us before. Most of the answers are public, but scattered across CSR filings, annual reports, funder websites and news. Sangam gathers them into one registry where every fact links to its source, and puts AI agents on top that answer questions, research new organisations and draft proposals without making anything up.

Two rules shape the whole system:

• Postgres is the system of record. Everything else, including the knowledge graph, is derived from it and can be rebuilt.
• Every fact carries a citation. An agent's answer cites the row behind each claim, such as [org:12] or [funder:4]. No answer comes from a model's own knowledge of an organisation.

Features

	What it does for you
Registry	Profiles of CSOs and funders: registrations, programmes, places, money given and received, news, each field with its source
Ask Sangam	Ask the registry anything in plain language. The agent writes its own read-only SQL, shows each step as it works, and cites every row it uses
Funder Match	Ranks funders for an organisation on seven criteria (theme, geography, population, eligibility, relationship, funding size, capacity), judged in code, with what would raise the score
Sangam Pitch	Drafts a proposal to a funder from facts on record and the files you choose. Every figure, place and name traces to a source or is refused, and it exports to Word
Sangam Scout	For staff and stewards: researches an organisation or a whole list from public sources, Sangam's library and your Doc Vault, and saves cited findings once a person approves
Media & Signals	News about the organisations and funders in the registry, classified and shown on their profiles
RFP Watch	Open calls for proposals from funders' own pages, matched to your organisation's themes

How it works

flowchart LR
  U["CSO fundraisers<br/>funder staff"] --> FE["Next.js app<br/><small>Registry · Ask · Pitch · Scout</small>"]
  FE -- REST --> API["FastAPI"]
  FE -- "AG-UI (SSE)" --> AG["LangGraph agents<br/><small>Ask · Scout · Pitch</small>"]
  API --> PG[("Postgres + pgvector<br/><small>system of record</small>")]
  AG -- "read-only SQL, as the caller" --> PG
  AG -.-> LLM["Claude<br/><small>Opus · Sonnet · Haiku</small>"]
  PG -- "derive" --> NEO[("Neo4j graph<br/><small>optional</small>")]
  style PG stroke:#0d6b60,stroke-width:2px

• The frontend is Next.js on Vercel. Sign-in is Supabase Auth, and the app's reads and Ask's SQL run as the signed-in person under Postgres row-level security.
• The agents are LangGraph deep agents that stream their tool calls to the screen, so you watch the work instead of trusting a black box. Ask Sangam has exactly six tools, four of them for SQL behind an allowlist, a validator and a read-only transaction (how).
• Models come in tiers by how hard the work is: Opus reasons, Sonnet researches, Haiku does fixed-shape work.

The architecture guide covers each subsystem and the decisions behind it.

Quickstart

You need Docker Desktop, an Anthropic API key and a free Supabase project.

git clone https://github.com/project-tech4dev/sangam.git && cd sangam
cp .env.example .env     # set ANTHROPIC_API_KEY and the two NEXT_PUBLIC_SUPABASE_* values
make up                  # start Postgres, migrate, then the backend and frontend
make seed                # load the cohort data
make smoke               # check the stack end to end

Open http://localhost:3000. The contributing guide covers configuration, tests, troubleshooting and how to open a pull request.

Built with

Next.js 16 · React 19 · FastAPI · Python 3.12 · LangGraph and deepagents · Claude · Postgres 17 with pgvector · Supabase · Neo4j and Graphiti (optional) · Langfuse · Vercel

Documentation

Doc	What it covers
Wiki	Generated reference: tools, routes, tables, settings, each claim linked to source
docs/ARCHITECTURE.md	How every subsystem works and why
docs/SQL_AGENT.md	The safe text-to-SQL design: catalog, validator, executor
docs/INGEST.md	Sangam Scout end to end: sources, matching, what it refuses to do
docs/DATA_PLATFORM.md	The data model and why it is shaped this way
docs/SKILLS.md	How the agents load procedures from SKILL.md files
docs/CHANGELOG.md	What shipped, by subsystem, and what is not finished
seed/data/README.md	Where every seeded record comes from

Glossary

Term	Meaning
CSO	Civil-society organisation, used interchangeably with NGO
CSR	Corporate social responsibility: Indian companies above a size must spend 2% of profit on it
Darpan	NGO Darpan, the government's registry of voluntary organisations; its ID is the closest to a national one
CSR-1	The MCA filing an organisation needs to receive CSR money; without it, it is ineligible
Schedule VII	The Companies Act list of activities CSR money may fund, and the top of Sangam's taxonomy

Status

Sangam is pre-release and under active development by Project Tech4Dev. The schema still moves, and master is what runs in production. docs/CHANGELOG.md lists what is not finished.

Contributing

Bug reports, ideas and pull requests are welcome. Start with the contributing guide, and report security problems privately as SECURITY.md describes. Everyone taking part agrees to the Code of Conduct.

License

MIT © 2026 Project Tech4Dev.