TechGuidr engagement · Economic research
Building two AI products from the ground up
How a large proprietary research archive and an existing commercial methodology became two practical AI products: a Research Intelligence Platform transforming research discovery and synthesis, and a Commercial Intelligence Workspace embedding AI into daily commercial execution. Product names are descriptive labels for this anonymised study.
The opportunity
An independent economic research business held a deep proprietary archive: over 78,500 publications, roughly 89,000 chart assets and around 35,700 structured forecast data points. Clients paid for the house view, but reaching it meant keyword search across an archive that size.
Ordinary “add AI” thinking was never going to work here. Answers had to reflect the firm’s current position, not a superseded one from 2011. They had to be grounded in the actual research, verifiable by the reader, and reliable enough to put in front of institutional clients. That is an architecture problem long before it is a model problem.
Product one: the Research Intelligence Platform
An institutional user asks a complex economic question in plain language and receives the organisation’s current house view, synthesised across the archive, with inline citations and the underlying research surfaced for verification. Materially beyond what keyword search could do.
- Hybrid vector and BM25 retrieval with semantic reranking
- Deliberate recency and supersession logic
- Structured forecast data injected alongside prose
- Multimodal processing of historical chart assets
- Separate classification, synthesis and grounding-validation stages
- Confidence indicators and source cards in the answer
Product two: the Commercial Intelligence Workspace
A commercial operating environment for account teams, with AI embedded in the actual working process rather than offered as an empty chat box.
Conceived, designed, architected and built by TechGuidr.
TechGuidr conceived, designed, architected and built both products from scratch: the product thinking, the architecture, the engineering, the data foundations, the AI implementation, the user experience and the delivery, all by one senior pair of hands, end to end.
The model was the easy part. The work was in the corpus engineering, the retrieval quality, the supersession logic, the deterministic commercial maths and the validation layer that made the outputs trustworthy enough to ship.
Found without being asked
While building the ingestion pipeline, Dan discovered that premium research assets intended to sit behind authenticated access were publicly reachable because of a content-delivery configuration problem. The firm had no idea.
It sat outside the AI product requirement entirely. He identified the issue, assessed the business risk and reported remediation to leadership. TechGuidr does not stop thinking at the boundary of the statement of work.
The outcome
The Research Intelligence Platform went into production. The Commercial Intelligence Workspace proved its value as a working prototype. A deep research archive became something an institutional client can interrogate in plain language and verify at source, and the commercial teams gained an operating environment with AI inside the workflow rather than beside it.
Have an AI use case too specific for off-the-shelf software?
This is what designing and building around your business looks like, rather than around a vendor’s demonstration.