The problem
Construction bids depend on reading blueprint sets and coordinating quantities and costs across trades.
← Selected work
Construction tech
Construction Estimation Platform
QPOI designed and built the AI analysis engine and core estimating workflow of a commercial construction platform, now in production with paying customers. It turns blueprint PDFs into estimates that teams can review and edit, then carries those figures into quotations, bid packages, contracts and project budgets.
- Client
- Confidential
- Platform
- Web
- Stack
- Angular / .NET
- Duration
- July 2025 to present
- Engagement
- AI engine and core platform development
- Status
- In production
The project at a glance
What QPOI delivered
QPOI designed and built the analysis engine, estimate review tools and connected project workflows, so reviewed figures carry through to quotes, contracts and budgets.
Delivery evidence
In production with paying customers. Our measurements showed median analysis time fall by roughly a third, from 24 minutes to 16 across 39 runs.
Document analysis pipeline, as builtView diagramHide diagram
1.0
The challenge
Preparing a construction bid means reading blueprint sets, calculating quantities and coordinating costs across multiple trades. The client needed to reduce that manual work while keeping the estimate open to review and correction.
The challenge extended beyond generating an initial estimate. When a reviewer changed a quantity, product or rate, the resulting quotes, contracts and budgets needed to stay aligned. Reviewers also needed to see which reports required regeneration.
2.0
The AI estimation engine
An AI-generated estimate is only the starting point. QPOI built the analysis engine and the review workflow around it, so teams can correct quantities and costs, restore earlier versions and carry reviewed figures into budgets and documents.
2.1
Blueprint to complete estimate
The pipeline analyses blueprint PDFs across 16 trades, producing material quantities, labour breakdowns and combined costs. Later stages prepare schedules, supporting reports and quotation documents for review.
2.2
Median analysis time cut by a third
QPOI redesigned the engine so independent trade analyses run concurrently and dependent stages follow in order. Our measurements showed median analysis time fall from 24 minutes to 16, comparing 19 runs of the previous engine with 20 of the redesigned version.
2.3
Deterministic numbers, AI at the edges
Reviewed quantities and costs live in a shared estimate store. Line-item edits recalculate totals in code and flow into the documents that use them. Specification changes carry into the figures and wording a contract is drafted from, and ambiguous text matches are skipped rather than risk changing the wrong item.
2.4
Human review built in
Reviewers can save and restore estimate versions, revert any line to the AI original, swap items for AI-suggested alternatives and see which reports an edit has made out of date. A refresh regenerates just those reports without rerunning the analysis.
3.0
Wider platform engineering
Working within the client's development team, QPOI took responsibility for the engine, core estimating workflow and the surrounding features below. The engagement combines new product development with ongoing improvements to an established Angular and .NET application.
Real-time infrastructure
SignalR / live progress / streamed AI
Designed and built the real-time layer for long-running work: live analysis progress that survives a page refresh, and the assistant's replies streamed as they're generated, so users watch long operations instead of waiting on them.
Performance engineering
Query profiling / request coalescing / Web Workers
Profiled and rebuilt the hot paths on both sides of the wire: a heavy calendar session went from 654 backend queries to 254, and loading the estimation screen from 53 API calls to 27. Request coalescing and caching on the frontend, identity-resolution caching and batch endpoints on the backend, and PDF report generation moved into a Web Worker.
Marketplace and geo search
Spatial data / Google Maps clustering / RxJS
Built the subcontractor marketplace: map-based search with distance filtering and clustered Google Maps rendering over live job postings, so trades find relevant work near them.
Team access
Invitations / roles / permissions
Designed and built team-based access: invitations, sign-up and roles matched to how construction companies actually operate, with per-member permissions over jobs, tasks and notes.
Bid packages and contracts
Trade packages / AI-drafted contracts
Per-trade bid packages generated from the estimate, trade award flows, and AI-drafted contractor-client contracts built from the reviewed figures instead of stale numbers.
Budget, timeline, and assistant
Budget sync / scheduling / AI chat
Budget tracking synced from the estimate, the project timeline and task scheduling, and an in-app AI assistant that answers questions in the context of the user's current project and stage.
4.0
Delivery and measured improvements
- Median analysis time cut by roughly a third in our measurements: 24 minutes to 16 across 39 runs
- Reviewed estimate figures carry through to quotes, contracts, budgets and reports
- The measured calendar session used 254 backend queries instead of 654; estimation-screen API calls fell from 53 to 27
- In production with paying customers: AI engine, project lifecycle, real-time and performance work
The work also keeps the product maintainable. QPOI upgraded the frontend from Angular 18 to 20 and split large application modules into focused components while preserving their public interfaces.
Behind the build / Engineering notes
How it works, in detail
The measurement methods behind these results, the pipeline stage by stage, and the faults caught during validation.
Built with
- .NET 8
- Angular 20
- Google Gemini
- Azure Cloud
- SignalR
- SQL Server
- Hangfire
- RxJS
- Web Workers
- Docker
- CI/CD Pipelines
Next project
Voice Sales Assistant