The Agentic Evolution Hackathon

Today, homeowners risk spending £5,000+ on architects just to assess feasibility (industry average, MHCLG / UK planning data) and then wait weeks to months for a decision, only to be rejected. PlanPass AI is an intelligent home planning compliance and design generation system for UK homeowners. Just enter your postcode and describe your plan. Behind the scenes, a team of AI agents analyses local council rules, checks compliance, and if needed automatically redesigns your layout into a fully compliant version. It even optimises key details like wiring, plumbing, and energy usage, and brings everything to life through interactive 2D and 3D visuals. The result is Instant clarity, smarter designs, and thousands saved, before you spend a penny on architects.

RunwayOps is an agentic payroll-risk command center for small businesses. It detects when cash timing puts payroll at risk, coordinates evidence from invoices, customer replies, bank events, and supplier terms, then recommends human-approved actions like emails, payment holds, and voice follow-ups. MongoDB stores the live case state, audit trail, agent runs, decisions, and memory so the system can adapt as new events arrive.

Volt Control is an agentic demand-response operator for renewable-heavy grids, built for the Agentic Retrieval track. It detects UK grid stress using live signals like frequency, demand, generation mix, weather, and interconnector flows, retrieves similar historical grid states, then identifies flexible loads such as EV chargers, HVAC, cold storage, and batteries to safely reduce demand. Low-risk actions are dispatched automatically, higher-risk actions require approval, and every decision and outcome is logged in MongoDB Atlas.

LiveRecall is hands-free clinical decision support for the bedside: a clinician points/asks a camera (phone or Ray-Ban Meta glasses) an unfamiliar medical equipment, asks a question, and gets a grounded answer with the authoritative text, thumbnail, and citation in seconds. LiveRecall ingests a multimodal catalog (text + image + vector embedding per apparatus) into MongoDB Atlas, uses Atlas Vector Search for semantic retrieval, and drives a multi-agent pipeline (vision → routing → retrieval → reranking → answer) coordinated via change streams-MongoDB as both the knowledge base and the agent bus. Decision support, not diagnosis-the clinician stays in the loop.

Substrate is a shared memory layer for AI agents working on engineering projects. It captures the work-in-flight context, drafts, decisions, claims, open questions, that lives between commits and never makes it into GitHub, Slack, or a CLAUDE.md, so agents working in parallel converge through shared working memory instead of diverging on merge. Built on MongoDB Atlas with $graphLookup and Vector Search composed in a single aggregation, exposed to any agent via MCP.

Housing data aggregator for building services. Analyses public data (land registry/planning permissions) as well as user complaint submission. Process the data using AI. This allows regulating bodies or councils or even home buyers, using a chatbot to ask question about the data. Aggregated reports or specific building information can be queried