Inspiration We saw a ₹500 crore construction project halt for a month just because a custom electrical panel was delayed. The project manager had no idea until it was too late because they were tracking materials manually across Excel sheets and WhatsApp. We realized that scheduling tools are completely blind to material deliveries. We built ConstructIQ to connect the project schedule with real-time supply chain updates.

What it does ConstructIQ is an AI "brain" for construction projects: Reads Documents: Extracts details from purchase orders, bills of quantities, and specifications automatically. Predicts Delays: Tells managers if a material will be late weeks in advance. Reconciles Deliveries: At the site gate, workers snap a photo of the delivery ticket, and the AI instantly matches it against the order to flag errors. WhatsApp Assistant: Site crews can check material statuses in Hindi or English via WhatsApp.

How we built it We combined several modern open-source AI tools: Ingestion: Qwen2.5-VL and Docling to extract tables and text from PDFs and drawings. Knowledge: Microsoft GraphRAG to link contracts and specification documents. Forecasting: PyTorch GNNs and time-series models to calculate delay probabilities. Automation: LangGraph state machines to write emails and draft purchase changes. Interface: React dashboard and a Twilio-powered WhatsApp bot.

Challenges we ran into Messy Drawings: Standard scanner tools couldn't read blueprints with rotated text and tables. We resolved this by using Vision-Language AI that reads the drawings like a human engineer. Limited Data: Construction sites do not keep neat digital logs. We solved this by teaching our AI basic logistics rules (like truck capacities and transport routes) to improve predictions. Long Procurement Cycles: Procurement takes months, but AI agents usually lose memory quickly. We resolved this by saving the AI's state in a database so it remembers its tasks over time.

Accomplishments that we're proud of Schedule-to-Material Link: Built the first system that automatically reschedules tasks when a material fabrication is delayed. High Accuracy: Reached 93.3% delay prediction accuracy on limited site data. Simple Field Bot: Created a multilingual WhatsApp interface that site workers can use easily on their phones.

What we learned Keep Site Work Simple: Field workers will not fill out complex forms. Taking a photo at the gate is the only thing that works. Delays Happen Early: Most delays start in the drawing approval phase, not during transport. Basic Search Fails: Technical specs are too complex for simple keyword searches; they require relational graph indexing.

What's next for ConstructIQ ERP Connectors: Link directly to accounting and planning tools (SAP, Primavera P6). Site Cameras: Use drone and site camera feeds to check progress automatically. Supplier Trust Registry: Build a database ranking suppliers by on-time delivery across the industry.