This resource is powered by Insight Bridge
Insight Bridge is a pipeline that equips you to wield knowledge you didn’t know you needed, by giving it a structure any question can travel through.
Evidence that could change how something is done goes unused for two reasons: there is too much of it to get across, and reaching it means already speaking the language of the body that produced it. Both costs fall hardest on small organisations, community groups and individual advocates, who have no research staff to absorb them.
Insight Bridge opens a body of evidence to whoever has a question, and makes what is inside it findable in the words you would actually use.
What Insight Bridge is built to respect
Three commitments shaped how this pipeline was built. Each one is about whose knowledge counts, and who gets to use it.
- The detail. Public purpose work depends on a clear line of sight to what was actually said, and by whom. Every proposition traces back to the passages and the verbatim quotes it was built from, and every source links out to the original.
- The minority. How common a view is in a corpus does not automatically say how much it matters. Whether you want the position of a specific community or a specialist perspective is a judgement only you can make. Cutting the clustering at its leaves keeps each small distinct concern as its own topic.
- Your judgement. You bring your own context, priorities, and principles. Insight Bridge equips you to discover what is argued, who argues it, and the specific points they raised. It does not rank sources, score their credibility, or tell you who is right.
AI Agents + Insight Bridge = Democratisation
Speak · Surface · Synthesise
Insight Bridge does the structuring. An agent does the reading, the matching and the interpreting. Between them they turn what a person cares about into something the record can answer, then turn the answer back into something that person can use.
- Speak
Say what you care about, in your own language: the thing you have spent a career on, the situation you are actually in, what you are trying to change. No one should have to learn a corpus’s vocabulary to get an answer from it.
- Surface
An agent translates that into the record’s own terms and brings up what speaks to it, across the whole body of evidence, including material you could not have named, written for a purpose that was not yours. The structure is what lets it do that without reading four thousand documents itself: it searches the analysis, walks the topic tree, and pulls the passages and quotes sitting behind anything it finds.
- Synthesise
The agent then reads what it surfaced against the context you gave it in the first step. It works out which findings bear on what you are trying to do, organises them around your problem, and writes the result in the terms you arrived with.
Real world example
An advocate specialising in Australia’s mental health sector, who also cares deeply about child safety, gendered violence and education, described all of that in their own words, with no idea how any of it was filed in the Watchful State corpus. What came back was:
- Six parts, following the shape of what they had described: mental health, then child safety and shame, then gendered violence, then education, then the machinery that let all of it persist, then what to do about it.
- Thirteen findings, each argued from reports it named, quoted and dated, with the oversight body that made them and its severity score attached.
- Eight recommendations, every one of which an oversight body had already put on the record.
- Connections between things filed apart: that a family violence response is suicide prevention, and that exclusion from education and untreated mental illness feed each other.
- An evidence ledger of the twenty-three reports it drew on, and a note on how it was assembled.
Whatever field you work in, the evidence that would change how you work has probably already been gathered by a body you have never heard of, for a purpose that was not yours.
Inside the Insight Bridge pipeline
How dense information becomes accessible
A body of documents becomes reachable when it has structure, and these seven stages are how it gets built: every document read in full, every passage placed among the passages arguing the same thing, and every source given a position on each of those arguments. They run the same way whichever corpus goes through them.
- 01
Curation
Sources are collected and normalised into documents. This is a hand-built evidence base rather than a sampling frame, and the selection is the first and largest analytical choice in the pipeline. Counts therefore describe what was gathered, never what is common in the world.
- 02
Facets
Each source is classified on the axes that matter for the corpus (who is speaking, when, what kind of body or publication) by deterministic rules over its metadata; the model plays no part in it. These become the comparative lenses the app offers.
- 03
Extraction
Every document is read end-to-end by a language model, which pulls out its key points with verbatim supporting quotes, tags it against controlled vocabularies, and records structured analyses such as the claims it makes and the evidence offered for each.
- 04
Clustering
Passages are embedded and clustered from the bottom up, and the resulting hierarchy is cut at its leaves: the smallest coherent groupings survive instead of being absorbed into the broad, stable ones above them. A distance penalty pushes a single source’s own passages apart, so a topic has to be reached by several sources. The topic list is an output of this stage; nothing supplies it beforehand.
- 05
Membership
Each topic then works out which dimensions of meaning actually separate it from the rest of the corpus, and grades every passage on that topic’s own measure. A passage is an exemplar of a topic, a high-value member, or a member, so the strongest evidence for a topic can be told apart from its edges.
- 06
Perspectives
Each cluster’s proposition is synthesised from its members first. Only then is every engaged source assessed against that completed proposition, giving a position and the reasoning behind it. The same is done per lens value, producing the comparative views.
- 07
Grouping
The topics are clustered again by the same method, this time on a representative vector for each, producing themes and then families above them, until the top generation is small enough to hold in your head. Membership stays soft at every level, so a topic can belong to more than one theme, with one marked primary. What you get is a tree you can walk down and cross-links you can follow sideways.
Other corpora on this pipeline
The same pipeline, the same reading conventions, different bodies of evidence. Each is a separate app with its own corpus and its own MCP server. Because the structure and the reading rules are identical across all of them, a question that starts in one can be carried into another without learning anything new.
Every corpus is available to agents
Each Insight Bridge app exposes its corpus over the Model Context Protocol, so an agent can work the analysis directly. The shape is consistent across corpora: orient on the whole body, search passages semantically, walk the topic map, open a topic to see its proposition and who stands where, split any topic by a comparative lens, and pull the record for a single source.
The tools carry the reading rules in their own descriptions, so an agent is told, at the point of use, that counts are not prevalence and that positions are relative to a framing.