Work
Three case studies: a production multi-agent platform we built and operate as our own product, enterprise AI shipped on Databricks at a global events group, and operational analytics that paid for itself.
Case study · Our product
BuildGuider — AI agent employees for residential construction
The problem. Small construction businesses run on scattered signals — WhatsApp messages, voice notes from site, emailed drawings, paper invoices. Estimating is slow and error-prone, scope drifts silently, and nobody can reconstruct why a decision was made.
What we built. A production platform where an orchestrator and seven specialist agents (estimator, quantity surveyor, procurement, contracts, site, commercial) work from one governed business memory:
- Own agent runtime — streaming responses, per-conversation sessions, a typed tool registry with permission classes, and human approval-and-resume flows for consequential actions
- Knowledge-graph memory — the graph is the single source of truth, with versioned memory projections, trust scoring and decay, and a background curator that reconciles new facts
- Omnichannel intake — WhatsApp, SMS, email, voice notes and chat unified through a triage agent, with identity verification and provenance-cited replies
- Document intelligence — PDF drawing extraction, invoice pipelines and voice transcription feeding a deduplicated document registry
- Multi-tenant security — every path to ground truth is a tenant-stamped, validated, audited tool; an append-only activity ledger doubles as the audit trail
Impact. Pre-estimate turnaround down 80%, estimating cost down 70%, scope completeness up 30% — in production with live users. Stack: Claude managed agents, Python, Next.js/TypeScript, TiDB (+ vector), R2, Docker.
Case study · Enterprise
Production AI on governed data at a global events group
The context. A global events business — 20+ business units, 120+ live events, 1,000+ digital products — with customer data fragmented across five CRM and event platforms, and no governed foundation for AI.
What we did. Built the data foundations first, then shipped the organisation’s first production AI agents on top of them:
- Genie domain-specific agents in Databricks, shipped to production — governed natural-language analytics in the hands of business teams
- Curated gold layer architected for both agentic and analytics consumers, with semantic layers pulled back from Power BI into Databricks so every consumer computes from the same governed logic
- Entity resolution across five sources (deterministic + probabilistic matching): duplicates down 32%, cross-platform matching up 45% — now scaling to 30m+ profiles from 18 sources
- Governed Reference Hub in Databricks Apps for reference data and data quality, replacing spreadsheet-managed lookups
- Enterprise ontologies, knowledge graphs and metadata standards, with governance migrated to Unity Catalog — data quality incidents down 24%
Impact. £50m+ of revenue decisions informed and £30m+ of churn revenue recovered through unified customer data — plus the group’s first three-year AI strategy, which we co-authored and piloted.
Case study · Operations
Operational analytics & ML that paid for itself
Manufacturing. For a UK manufacturer: Azure/Databricks pipelines supporting ERP, manufacturing and logistics; a full migration from legacy Tableau to Power BI; early copilot-style analysis over warehouse data; and exception reporting that lifted process efficiency ~12% and cut data inconsistencies ~18%.
Construction, fuel logistics & agriculture. Across independent consulting engagements: £485k of cost reduction through supplier consolidation and data-driven negotiation; fraud and anomaly detection models over fuel usage and machine hours; and yield, pricing and forecasting models delivering 7–17% revenue uplift.
The common thread: analytics built inside real operations, adopted by non-technical users, and measured against commercial outcomes — the same discipline we now bring to AI systems.
Thinking behind the work
Why does construction estimating go wrong so often — and so expensively? Read our essay: The Cost Reality Gap in Residential Renovation.
DI Consultants 