Introducing Cortex
Cortex turns files and documents into a persistent, continuously updated knowledge base that supports a range of uses, from management information (MI) and board reporting to regulatory analysis, risk monitoring, policy research and on-demand dashboards.
In 2023, I gave a live demo in a webinar on generative AI. I fed my old blog posts into a retrieval-augmented generation (RAG) pipeline and asked questions in real time.
AI-enabled knowledge bases have come a long way since then. Cortex is my take on what they should do today.
Cortex reads your files and documents, builds a connected knowledge base, identifies questions the documents leave unanswered, and among many other use cases, creates dashboards on demand.
Lessons from the first attempt
Cortex is not my first attempt at an AI-enabled knowledge base. In a previous role, I led a project to build an internal IFRS 17 knowledge base. It didn’t work well enough; the answers were unreliable, and the cost was hard to justify.
The problem was not simply the choice of model. It was the approach. RAG retrieves relevant passages when you ask a question. The model uses those passages to construct an answer. This works for many questions, but it can struggle when the answer depends on relationships across several documents, or on details that are easy to miss in a short extract.
Cortex takes a different approach. Rather than starting from scratch with a handful of passages for every question, it reads documents and builds a persistent, interconnected knowledge base.
It creates pages for important entities and concepts, records figures alongside their page references, and flags contradictions between sources. When a new document arrives, Cortex updates the relevant parts of the knowledge base.
What Cortex does
- Reads and connects documents. Give Cortex PDFs, financial statements, meeting minutes, policy documents or web links. It extracts the content, records the important information with source references, and connects related facts across documents. Figures are organised into sourced tables that can be used in reports, models and dashboards.
- Finds unanswered questions. Cortex tracks what the documents establish and what they leave unclear. It flags gaps, conflicting statements and figures that do not reconcile, giving the people responsible for the data a list of questions to investigate.
- Builds dashboards from the evidence. Ask Cortex for a dashboard and it generates a self-contained page in your preferred style, drawing on the knowledge base. Figures link to the documents and pages they came from, so readers can check the underlying evidence.
Putting it to the test
I gave Cortex the public record of the Institute and Faculty of Actuaries (IFoA): financial statements, Companies House filings for its subsidiaries, Civica reports on Council elections, and Council and Board minutes. Members, like myself, have asked for clear information about membership, examinations, finances and governance.
While not perfect (and there is a long list of disclosure gaps), there is sufficient publicly available information to build a useful IFoA transparency dashboard. This is an independent demonstration built entirely from public sources. It is not an IFoA publication, and the work relied only on publicly available information.
Who it’s for
Cortex is for institutions that rely on large collections of documents: banks, insurers, pension schemes, asset managers etc.
A reporting team could use it to trace figures back to source documents before signing off a report. A board secretary could use it to prepare for the questions directors are likely to ask. A new joiner could search a policy library without having to read every document from cover to cover.
The common problem is the same: the information already exists, but finding it, connecting it and checking it takes time.
Get in touch
Cortex is open for pilot projects. If your organisation has a document problem worth solving, email [email protected] or book a call. Bring a real use case; we’ll see what Cortex can do with it.
The bottom line: finding fragments is not the same as understanding the bigger picture. Cortex builds a connected knowledge base from your documents, keeps it updated as new information arrives, and traces every answer back to its source.