Description
The client, a US-based construction technology stakeholder, is building an AI-driven system that centralizes and interprets large-scale building documentation. The platform connects to Microsoft SharePoint repositories containing architectural drawings, electrical schematics, HVAC layouts, and maintenance manuals for complex buildings such as airports and commercial infrastructure. The system uses AI models (via Google Gemini API) to extract, interpret, and contextualize information from structured and unstructured documents, enabling natural language querying across massive engineering datasets.
Challenge
The project was initially delivered by an external developer who significantly underestimated both scope and complexity, leading to a major deviation from the agreed timeline and expected outcomes. Although the original estimate was six weeks, development extended beyond ten weeks without delivering a complete or production-ready system. As a result, the client was left with an incomplete product and uncertainty regarding how much of the initial scope had actually been implemented.
An independent technical review was required to objectively assess the state of the project, validate completion against the original requirements, and determine whether the implemented features matched the intended scope. During the audit, it became clear that the codebase had evolved in an unstructured way, with signs of AI-generated and inconsistent legacy-like code, unnecessary features that were not part of the original scope, and missing core SaaS fundamentals such as proper organizational structure.
The client also needed clarity on whether the system could be salvaged and how far it was from a usable MVP. Overall, the challenge was to bring transparency to a heavily misaligned project, separate real implemented value from noise, and define a clear path forward from an unstable, partially built system to a structured, scalable AI-powered platform.
Solution
We performed a full technical and functional audit of the platform:
- Extracted and structured the full original scope into atomic feature units
- Mapped each feature against the existing codebase
- Assigned completion percentages per feature based on implementation status
- Conducted deep codebase analysis to identify architectural and AI design issues
- Delivered a formal report confirming scope deviation and system maturity (~80% functional alignment on core features, but low architectural correctness)
After validation, we:
- Accepted the project from the previous developer
- Re-architected the system toward a proper SaaS model (including organization-level structure)
- Defined and implemented Phase 1 MVP completion requirements
- Designed Phase 2 architecture for real-time data integration
- Improved AI retrieval quality using curated “golden queries” (training question-answer pairs)
- Optimized document understanding pipeline for engineering drawings and manuals
Technology Stack
- Frontend: Next.js, React
- Backend: Python
- AI Integrations: Google Gemini API, document parsing + retrieval-augmented generation (RAG)
- Data Layer: Microsoft SharePoint integration for document storage and retrieval
- Other: SaaS architecture design, AI-driven document understanding, knowledge graph-like retrieval system, real-time IoT sensor integration (Phase 2)