How the base is built, and how an answer and a fit score are made.
What the base is
Sangam is three layers, each built on the one below. Nothing in a higher layer is allowed to state a fact the data layer does not hold.
Where the data comes from
Every row records its source URL, the source's name and when it was fetched, so any field can be traced back to where it came from.
- NGO Darpan
- The government registry of NGOs: registration, credentials such as 12A, 80G and FCRA, office bearers and working sectors.
- Funder filings and publications
- CSR reports, annual reports and statutory disclosures, read by Sangam Scout from sources whose terms allow it.
- News and signals
- News coverage of organisations and funders, collected and held for review before it is shown.
- Organisations' own documents
- What an organisation uploads to its Doc vault. Private to that organisation unless it chooses otherwise.
From source to Registry
How an answer is made
Ask Sangam is an agent that reaches the base only through read-only database tools. For each question it:
- Reads the question and works out which organisations, funders or scores it is about.
- Checks which tables hold that information, then writes a read-only query against them.
- Answers from the rows that come back, putting a citation next to every fact; each one links to its Registry entry.
- Says so when the base has no answer, instead of filling the gap from general knowledge.
Every answer shows these steps above it, so you can see which tables were read and why. It reads a web page only when you ask it to, and then says where the text came from.
How a fit score is made
A versioned rubric — v1 has seven criteria worth 100 points; v2 splits the same criteria into weighted sub-criteria — is applied to each organisation and funder pair. Every criterion score carries a written justification, and any manual correction is kept as an adjustment with its reason, so the audit trail ships with the score.