A conversational librarian holding the collective knowledge of India’s social sector
Every knowledge resource this sector creates deserves to be found by the person who needs it — at the moment they need it & in the language they read.
The page of Google results past which almost nobody looks. Below it sit the several thousand organisations working on your exact problem statement.
Project brief — on why search-engine discovery structurally fails a sector this large
India’s development ecosystem produces an extraordinary amount of knowledge. Baseline surveys, impact evaluations, case studies, field toolkits, M&E templates, annual reports. Some of it costs a crore to produce. Almost all of it is written by people who learned something on the ground that nobody else knows yet.
And then it is put on a website, announced once, and left there.
This tool does not try to make that knowledge better. It is already good. It tries to make it findable — which turns out to be the thing standing between the knowledge and its entire purpose.
Nothing here is anyone’s fault. Every organisation is behaving rationally. The system they are behaving rationally inside is the problem.
An organisation finishes a two-year study. It goes up on their website, because that is where an organisation’s knowledge goes. They write a LinkedIn post with a link. They email their newsletter list. If the budget allows and the study warrants it, they convene a roundtable.
That is the entire outreach a piece of knowledge gets. And notice what it depends on: whether you already follow that organisation, and whether you happened to need that specific thing during the week they posted about it.
Both conditions fail almost always. You cannot follow every organisation — there are more than thirty lakh non-profits in this country. And need does not arrive on a publishing schedule. You need the horticulture M&E template in March, and it was announced in September, and you were busy in September.
So the sector falls back on search engines, which means it falls back on SEO budgets, which means the organisations best at being found are the ones who could afford to be found. And even then only the first three pages exist, in a sector with thousands of organisations per domain.
Which leaves a quieter workaround that many of us actually use: a private list of organisation websites you check by hand, every few weeks, hoping something new is up. Sometimes there is. Usually there isn’t. That is a search engine made of a person’s free time, and it is what the sector’s knowledge discovery currently runs on.
“Every now and then I discover a really good organisation with impressive work, and it is disheartening to see good knowledge resources parked on their website, knowing their limited reach.”
From the project brief
What a single large study can cost to produce — before the question of whether anyone outside its funder ever reads it.
Project brief, on the cost of sector research at scale
How often the same question gets researched when the first answer was never findable. This is the real bill: not the study, the second study.
Implied by the brief’s “build beyond, not again” principle
What limited reach actually costs. Documentation that nobody reads stops looking like an investment and starts looking like overhead — so it gets cut first.
Project brief, on how low reach discourages knowledge creation
Every tool in DILSE takes one vicious cycle the sector already runs, and changes a single step in it. Here is this one.
Same five steps. Same shape. One of them changes.
An organisation learns something real on the ground and writes it up.
It is published to their own website — one address among tens of thousands.
Reach depends on who already follows them, on the day they post.
A few hundred reads. Most of the sector never learns it exists.
Documentation looks like overhead. The next study shrinks, or never happens.
An organisation learns something real on the ground and writes it up.
It is published to their own website — one address among tens of thousands.
Reach depends on who already follows them, on the day they post.
A few hundred reads. Most of the sector never learns it exists.
Documentation looks like overhead. The next study shrinks, or never happens.
Stop announcing the resource. Start being asked for it.
Everything about the current system is a broadcast: a post, an email, an event. Broadcast has two structural dependencies that no amount of effort can remove — the size of your network, and the coincidence of your timing with someone else’s need.
Retrieval has neither. If the resource is in a shared index, the person who needs it does not have to have heard of you, and does not have to have been paying attention in September. They ask in March, and it is there.
That is the whole change. One upload replaces an outreach campaign, and it keeps working after the campaign would have ended. The organisation’s effort drops to near zero and its reach becomes permanent rather than momentary.
The second-order effect matters more than the first. Once reach is visible and durable, documenting stops reading as a cost centre. The thing that most limits knowledge creation in this sector is not curiosity or capability — it is the well-founded suspicion that nobody will read it.
The same act — writing up what you learned — lands completely differently on both sides of the flip.
Searches, finds page three, gives up, and designs from scratch — or asks a WhatsApp group and hopes someone happens to know.
Asks in plain language and gets five case studies from five states, each attributed, each linked, each with the option to interrogate the document directly.
Reach is capped by follower count. A genuinely useful toolkit reaches a few hundred people and stops.
The toolkit is discoverable to the whole network from the minute it is uploaded — and stays discoverable, with the analytics to prove it.
Documents in their own language and accepts that the sector will not read it. Or writes in English and loses precision.
Documents in their own language and is read anyway — the index does not privilege the language a resource happens to be in.
Receives the PDF. Has no way to know whether it informed anyone’s decision but their own.
Can see the resource being found, by whom, and how often — the outcome the research was funded for in the first place.
Concludes it doesn’t exist, and re-researches it. Nobody else ever learns the gap was there.
Files a knowledge request. The gap becomes a visible, countable signal that the sector is missing something.
Everything above is the what and the why. The demo is the how — so here is where each claim goes to be checked.
Discovery-first, never answer-first — the librarian surfaces the resource, it does not replace reading it.
Language is not a filter on what you can find.
Once a resource is surfaced, you can interrogate it without reading 200 pages.
Uploading is the entire dissemination effort.
Reach becomes visible, which is what makes documenting fundable.
When nothing exists, the gap becomes a signal instead of a dead end.
This is not a speculative technology. Every component of it is running somewhere at national scale already.
Standard, well-understood practice: the model answers only from indexed documents and cites them.
What it provesThe hard part of this tool is not the AI. It is getting organisations to upload — which is a design and trust problem, not a technical one.
India’s national school education platform, built on Sunbird’s open building blocks, serving content at population scale.
What it provesA shared national content index with many independent contributing institutions is a solved architecture in this country.
The handful of sector platforms that did reach real adoption were the ones that aggregated rather than published.
What it provesWhere discovery was solved by design, adoption followed. Where each platform fought for visibility alone, it did not.
A searchable index is the visible half. The half that matters more only shows up after a few years of it running.
Every unanswered knowledge request is a datapoint about what the sector does not know. Aggregate a few thousand of them and you have something no ecosystem currently has: a live, evidence-based map of its own blind spots. That is a research agenda written by the field instead of for it.
The most valuable knowledge in this sector is not in reports at all — it is in what an organisation in Jodhpur learned by trying something that half-worked. Make that findable and organisations start learning from each other rather than from consultants.
A meaningful share of sector research re-establishes things already established elsewhere, because the first study could not be found. Every one of those avoided is a study’s worth of money and two years of talent redirected to a genuinely open question.
This is the one that changes behaviour. When an organisation can show that its last field toolkit was found and used four hundred times by people it has never met, the internal argument for documenting the next one is already won.
The intent behind funding a study is never the PDF. It is that the ecosystem becomes better informed and decides better. Everyone involved — the funder, the researcher, the field team who gave their time to the survey — was working towards that outcome.
Discoverability is the step where that intent currently fails. Not rigour, not writing, not effort. Findability.
Which is worth sitting with, because it means the fix is unusually cheap. The knowledge is already produced and already paid for. What is missing is an index — and an index is the least expensive thing in this entire story.
DILSE is one infrastructure, not ten tools. These are the loops this one feeds, and is fed by.
Every project profile on the map carries its own documents — baseline surveys, evaluations, case studies. Those are library resources. Find a project you want to learn from on the map, and its evidence is already indexed here.
The Forum answers questions the library cannot: local, tacit, unwritten. When a forum thread keeps re-asking the same thing, that is the library telling you a resource is missing.
A course is knowledge someone took the trouble to structure and teach. The library is where the reading behind it lives, and where a learner goes deeper after the module ends.