Knowledge Engine
Our role: Productionization architect: the reusable core, NFRs & security.
The Challenge
The sales team couldn't answer detailed product questions on the spot — reading full manuals mid-conversation wasn't practical. An R&D analyst had built a promising chatbot prototype the sales team already found useful in pilot, but it lived in a notebook: not production-ready, not secured, and not reusable.
Solution & Approach
We took the analyst's prototype, defined the requirements and non-functional needs, and architected it into a production system on Azure — directing the vendor team that built it. The defining decision was generality: rather than a point sales tool, we designed a modular product-knowledge engine with a reusable core (chat UI, streaming, history, chunking, search, logging, file handling, shared authz) over clean integrations (Azure Document Intelligence, SharePoint, OpenAI, Blob, Key Vault). That core then served three uses from one codebase — the sales chatbot, ESG tender and compliance drafting, and a product-deck generator using hierarchical prompting. We owned the security, including backend validation of every auth token against the identity provider's certificate, and solved the real RAG wall — clean text out of image-heavy and footnoted-table documents — with an OCR-backed parsing pipeline.
The hard call — build the engine, not the feature. The easy path was a point tool for sales. We invested in a generic, modular core instead — and it paid off: the same codebase went on to answer ESG tenders and generate product decks, three uses for the price of one.
- One reusable core — Chat UI, streaming, history, chunking, search, logging, and file handling, shared across every use.
- Source-cited answers — Every reply opens to the exact page of the source document, so a rep can verify before repeating it.
- Automatic, current index — New documents flow from SharePoint through Document Intelligence into the search store on a schedule — no manual ingestion.
- Reused three ways — The same engine powers the sales chatbot, ESG tender drafting, and product-deck generation.
Outcome
In production use and adopted by the sales team. All product documentation is indexed and kept current automatically, every answer ships with a source citation that opens to the exact page for verification, and answers return instantly. Because the core was built to be reused, the same system went on to draft ESG tender sections and generate product decks — three applications from one codebase.
The long version: this build has a field note in our journal — the decisions, dead ends, and judgment calls behind it. Build the engine, not the feature →
Let's discuss your next challenge.
Ready to bring AI-powered intelligence to your organisation?
Start the Conversation