AI-Assisted Intelligence Feed and Knowledge Sharing for an Advisory Business

Advisory, consulting and knowledge-led organisations that depend on staying ahead of their market and sharing what they learn. This case is most useful if:

• your people follow dozens of sources but nobody can say which articles matter
• insight stays in individual inboxes instead of reaching colleagues and clients
• you want AI to filter and draft, but not to publish without a human
• AI running costs have to be predictable and capped

The client relied on a stream of industry news, analyst blogs and newsletters to inform its advisory work. Reading was manual, relevance was a matter of memory, and the useful findings rarely turned into anything colleagues or clients could see.

It needed a feed that knew what mattered to the business and helped turn reading into shareable insight, without handing editorial judgement to a model.

We built an AI-assisted intelligence system that runs from subscribed sources to publication-ready drafts, with a person approving every step that leaves the system.

• Sources and reader: OPML import from the client's existing feed reader, single-feed onboarding with a health check, and a three-column reader with saved items and full-text search.
• Radars: user-defined interest profiles that score every incoming article for relevance, with a ranked "For You" view and opt-in discovery of new sources through the GDELT news index.
• AI commentary: one click turns a selected article into three LinkedIn drafts of 150 to 250 words, each written through a different analytical lens and calibrated against a library of the client's own best posts.
• AI blog pipeline: Claude pre-scores candidate topics on six dimensions (minimum 18 of 30), sources supporting evidence through web search and suggests internal links. Nothing is published automatically.
• Evidence and provenance: Postgres as the single source of truth, immutable evidence records, and every AI output stamped with what produced it.
• Cost controls: hard spending ceilings on AI use, such as 3 US dollars a month and 5 cents a request for commentary, with a twice-daily background worker.

Relevance by design, not by memory. Radars rank what arrives against the topics the business cares about, so the important article no longer depends on someone happening to read it.

From reading to sharing in one click. AI drafts in the organisation's own voice, written for a Nordic audience that often reads English as a second language, turn private reading into material colleagues and clients can use.

AI with a human editor and a budget. Every draft is reviewed before it leaves the system, every output can be traced to its evidence, and running costs are capped by design.

Most intelligence tools add more to read. The useful ones decide what not to read, and make the rest easier to pass on.