ProBuild: An AI Operating System for Construction
How QPOI engineered a proprietary "Blueprint-to-Timeline" engine that transforms static PDFs into live, intelligent project schedules using multi-stage LLM orchestration.
Duration
6 Months
Platforms
Web (Angular + .NET)
Tech Stack
Angular, .NET Core, SQL Server
Engagement
End-to-End Architecture
The "Unstructured Data" Problem
Construction projects start as dense, 100-page PDF blueprints. Extracting data from them—like concrete volume or electrical timelines—is a manual process that typically takes estimators days of work.
Our client needed a system that could ingest these technical documents, understand complex architectural context without hallucinating, and output structured financial estimates in minutes.
QPOI's Role
- End-to-End System Architecture
- Complex AI Pipeline Orchestration
- Real-time Infrastructure Implementation
The "Blueprint-to-Timeline" Engine
We engineered a sophisticated ingestion pipeline that treats a PDF not as a file, but as a dataset waiting to be unlocked.
Sequential Prompt Chaining
Instead of a single AI query, we built a multi-phase backend process:
Phase 1: OCR extraction & structural analysis.
Phase 2: Identifying foundation & wall systems.
Phase 3: Generating & sequencing construction tasks based on industry standards.
"Self-Healing" Workflow
AI isn't perfect, so we designed for ambiguity. If the model finds conflicting data (e.g., slab vs. crawlspace), it triggers an interactive interview.
User corrections effectively "fork" the conversation, re-running subsequent prompts with the new truth, ensuring 100% data integrity.
Context Window Optimization
Construction projects generate massive context. We implemented a summarization engine that monitors token counts.
Before the window fills, it triggers a background summary prompt to condense history, allowing the AI to "remember" early decisions even 50 turns deep into the workflow.
Engineering Under the Hood
Real-Time Infrastructure
The app feels alive because of pervasive real-time features. We implemented two distinct SignalR hubs: ProgressHub for broadcasting high-level status updates across the team, and ChatHub for token-by-token streaming of AI responses.
Heavy Lifting on the Client
Generating a 50-page "Environmental Lifecycle Report" is computationally expensive. We offloaded this logic to Web Workers, ensuring the main UI thread remains buttery smooth while complex compilation happens in the background.
Geospatial Marketplace & "Find Work" Engine
Subcontractors needed to find relevant work without sifting through endless lists. We built a responsive, map-centric search engine that visualizes opportunities in real-time.
- Server-Side Clustering: Handling thousands of job pins using
MarkerClustererwithout performance degradation. - Dynamic Radius Filtering: Dragging a slider instantly updates the map overlay and result list via a dynamic spatial query system.
- Reactive Synchronization: Map and list views are synced via RxJS—hovering a list item bounces the corresponding map pin.
Enterprise Team Architecture
Construction companies aren't single users. We engineered a comprehensive self-service admin portal supporting complex hierarchy and Role-Based Access Control (RBAC).
- Invitation Loops: Secure, tokenized links allow admins to onboard employees directly into their workspace.
- Dynamic Seat Billing: Deep Stripe integration automatically tracks active users against purchased licenses, handling mid-cycle upgrades and proration logic.
Complex Financial Transactions
The platform needed to monetize transactions, not just subscriptions. We implemented backend logic to calculate dynamic "finder's fees" based on awarded bid amounts, automating the platform's revenue stream.
Business Impact
Reduced manual blueprint review from days to minutes.
Architecture scales to support heavy enterprise load.
Enabled SaaS subscriptions & transaction finder's fees.
Technology Stack
Need Engineering, Not Just Code?
We build the core systems that power complex businesses.
Let's Build Something Complex