Render Realty

Render Realty is an AI orchestration platform that transforms construction workflows, from 3D model generation from floor plans, HVAC and electrical layout planning, to seamless Solana payments.

Comment

Inspiration

Me and my teammates, were pondering industries where extensive optimization, due to the AI bubble, had not yet come in, and one of the industries we thought about was construction, specifically General Contracting.

What it does

Render Realty is an AI-powered orchestration platform for property owners, architects, engineers, and contractors. Key capabilities:

  • Auto-generate HVAC & electrical drawings from uploaded floorplans and deliver ready-to-download drawings for immediate use in planning and bids.

  • Convert 2D floorplans → 3D STL models for visualization in CAD viewers or 3D printing.

  • Find local trades (plumbers, electricians) via Google Places and surface the best-rated professionals with full contact details.

  • Estimate materials from floorplans (OCR + rules) with room-level breakdowns (studs, drywall, insulation, doors, windows, etc.).

  • Search real-time Canadian retail prices for construction materials (via GROQ-powered search) and return vendor-annotated pricing lists to simplify budgeting.

  • Create Solana payment links so contractors can send secure, shareable payment requests (SMS, email, chat).

How we built it

Backend & orchestration

  • Python-based orchestration layers coordinate image analysis (PIL + pytesseract OCR), material estimation logic, vendor price lookup, and external API calls.

  • A compliance engine uses a RAG approach (powered by Cerebras API) to check generated layouts against Ontario bylaws and flag likely violations.

  • Google Agent ADK acts as the root agent for workflow orchestration, delegating subtasks to specialist models and services.

  • Cohere Command A is used to manage control flows and high-level reasoning across the agent stack.

  • Groq serves AI search for material lookups and local vendor price aggregation using DynamoDB as it's backened.

  • Google Veo 3 produces short schematic videos that demonstrate proposed mechanical/electrical routing for stakeholder review.

Frontend

  • Lightweight React.js dashboard for chat-style interactions, role-based access control, and realtime reporting.

  • 3D model preview powered by three.js to render generated STL models in-browser.

Payments & notifications

  • Solana Pay integration to create on-chain payment links.

  • Privy for embedded wallet (onboarding), Phantom for user wallets, and Coinbase to transfer receipts/proceeds to a bank account everything culminates, in a frictionless "one-link" experience for the user.

  • Twilio for SMS notifications (payment requests, reminders, receipts).

Challenges we ran into

  • Different inference providers excel at different tasks (vision, fast retrieval, instruction-following), so we needed a robust orchestration layer to route each task to the right model/service (Cerebras/Groq/Cohere) and gracefully handle failures.

  • Building layout-generation algorithms that handle wildly non-standard floorplans while ensuring legal compliance was very difficult the variance in drawing conventions and missing metadata is a constant hurdle.

Accomplishments that we're proud of

  • Built a custom AI framework to generate reliable HVAC, plumbing, and electrical layouts from floorplans, improving consistency and output quality across diverse drawings.

  • Implemented end-to-end Solana payment flow (embedded wallets → Phantom transfers) with SMS delivery for payment requests, successfully tested in real transactions.

What we learned

  • Orchestration matters more than model choice. The hardest engineering problem was coordinating providers and building robust fallbacks, not model performance in isolation.

  • Standardization reduces error. Requiring minimal metadata (scale bar, known door/window sizes) dramatically improves downstream accuracy; otherwise the system must assume conservative defaults.

  • Compliance requires hybrid approaches. Combining deterministic rules + a RAG-backed reasoning layer (for edge cases) gives the best practical coverage for local code checks.

  • Contractors adopt tools that reduce friction, we need to build something that hooks them in and is automated downloads, one-click payment links, and clear material lists increase trust much faster than marginally better ML outputs.

What's next for Render Realty

Short-term (next 3 months)

Harden layout generation and compliance checks: extend the ruleset to cover more Ontario bylaws and add unit tests with real-world floorplans.

Improve floorplan preprocessing: better scale detection, vectorization, and heuristics for missing metadata.

Expand vendor coverage for Canadian material pricing (more retailers, better availability flags).

UX polish: simplify the contractor flow for creating and sending payment links and receipts.

Medium-term (3–9 months)

Pilot with local general contractors and an independent architect firm to gather real-world feedback and validate ROI (time saved, accuracy of takeoffs).

Add BIM/IFC export and tighter CAD integration so generated models can be used directly in industry-standard tools.

Add permit-check automation: surface probable permit needs and the documents required for common renovations.

Long-term

Marketplace & partnerships: integrate with supplier APIs and trades marketplaces so contractors can get instant quotes and schedule on-site visits.

Analytics & forecasting: add cost trend forecasting and scenario comparisons (e.g., alternate materials or suppliers).

Broaden compliance beyond Ontario: configurable regional rulesets and crowdsourced verification.

Built With

Share this project:

Updates