Our work
Recent projects
Case studies from our recent work. We can arrange references for buyers on request.
An AI assistant for tender submissions, with a person confirming every step
The challenge
A Welsh public-sector tender had 176 qualification questions across 10 sections, three envelopes, several mandatory documents and strict legal declarations. An accidental submission was not an option.
What we did
We built a secure web tool that drafts each answer and labels where it came from: public record, not applicable, proposed (please check), needs your input, or legal declaration (never pre-filled). Each section needs a review tick, each document needs a confirmation that resets if the file changes, and submission needs three confirmations tied to a fingerprint of exactly what was reviewed.
Built with Node.jsBrowser automationCloud RunIdentity-Aware ProxyCloud Storage
A working resident engagement platform, built during a bid
The challenge
A housing association wanted a platform for surveys, consultations, forums, events, moderation, consent and board reporting. It had to work for every resident and protect their personal data.
What we did
Rather than describe a solution, we built one that evaluators could click through: sign-in with MFA, consent history and data-rights requests, moderation, live dashboards that check who is taking part, board and regulator reports, exports, retention rules, SCIM provisioning, a documented REST API and an MCP server that gives AI tools pseudonymised data.
Built with Node.jsServer-rendered HTMLSQLite → Cloud SQLLive dashboards (SSE)MCPSCIM 2.0
Bid assessments from a full tender pack
The challenge
Tender packs arrive as a dozen or more PDFs, spreadsheets and Word documents, and they don't always agree with each other. Reading them properly takes days, and missing one clause can rule a bid out.
What we did
Our pipeline reads the whole pack and pulls out the timetable, scoring, mandatory requirements, insurance levels and evaluation rules. It cross-checks the documents, flags contradictions (such as a weekday printed differently in two places, or a pricing formula that leaves out rows) and writes a bid/no-bid assessment that cites its sources, with clarification questions ready to send.
Built with Frontier LLMsPDF/DOCX/XLSX parsingCitation tracing
Have a project in mind?
We can build a prototype for your problem.
Bring us a service, a backlog or a tender requirement, and we will show you what we could build.