AI-powered property defect analysis for the PropTech industry Built for the Taurgo × Cardiff University AI Hackathon — March 2026
© 2026 Sameer Shaik — non-commercial use only.
🎥 2-Minute Product Demo: Watch the demo
Property managers and landlords across the UK manage portfolios with hundreds of individual assets. Identifying, classifying, and prioritising structural defects traditionally requires a qualified surveyor to attend each site — a process that is expensive, slow, and difficult to scale.
Meanwhile, photographic records of properties already exist in abundance. The gap is not data; it is analysis.
SurveyAI closes that gap. Upload photographs of any property space, and SurveyAI generates a structured, RICS-aligned survey report in seconds — complete with defect classification, severity scoring, indicative repair costs, and prioritised action recommendations. Across multiple properties, the Portfolio Comparison Dashboard surfaces the most urgent issues across the entire estate in a single view.
No login required. Upload any property image (JPG, PNG, HEIC) to generate a report.
- AI Vision Analysis — Amazon Nova Pro on AWS Bedrock (eu-west-2) analyses property images and returns structured JSON conforming to a RICS-aligned defect schema
- UK Jurisdiction Compliant — All image processing stays within the London region; no data leaves UK/EU infrastructure
- Two-Step Prompt Chaining — Vision → structured JSON → professional RICS report using the same model
- Severity Classification — Five-tier system: Monitor / Low / Medium / High / Critical, with a 0–100 numeric severity score
- Defect Taxonomy — 8 Tier-1 categories and 40+ Tier-2 subtypes with confidence scoring per defect
- Defect Zone Overlay — Bounding box annotations on the source image showing the precise location of identified defects
- RICS-Standard Language — Survey descriptions, urgency ratings, and recommendations written to professional survey conventions
- Standards Citations — Relevant RICS guidance notes, British Standards, and UK Building Regulations cited per report
- Indicative Repair Costs — Low / mid / high estimates grounded in 2025 UK market rates (GBP)
- Risk Matrix — Likelihood × Impact scoring visualised as an interactive grid
- Analysis Limitations — Amber warning banner when image quality is insufficient for confident assessment
- Portfolio Comparison Dashboard — Severity timeline, defect frequency heatmap, and combined cost estimate across all analysed properties (activates at 2+ images)
- Priority Action Summary — Surfaces the single most urgent defect across the portfolio, ranked by severity first, then frequency and confidence
- Report Q&A — Ask natural language questions about any individual report; answered in context by Amazon Nova Pro on AWS Bedrock
- PDF Export — Full multi-page PDF report with cover page, per-property report pages, and portfolio summary page (jsPDF)
- Copy to Clipboard — Full plain-text report for pasting into existing property management systems
- HEIC Support — iOS HEIC images converted to JPEG automatically on upload
- Session Caching — Fingerprint-based caching ensures re-uploading the same image returns a consistent result within a session
- Analysis Progress Modal — Per-image stage tracking (preprocessing → classifying → generating → scoring)
┌─────────────────────────────────────────────────────────────────┐
│ Browser (Client) │
│ │
│ ┌─────────────┐ ┌──────────────┐ ┌───────────────────┐ │
│ │UploadSection│ │useImageAnalys│ │ ReportSection │ │
│ │ │───▶│ is │───▶│ ComparisonDash │ │
│ │ File picker │ │ (React hook) │ │ ReportQA │ │
│ │ HEIC→JPEG │ │ imagesRef │ │ DefectHighlight │ │
│ │ Fingerprint │ │ Session cache│ │ Viewer │ │
│ └─────────────┘ └──────┬───────┘ └───────────────────┘ │
│ │ │
│ fetch POST /bedrock-analyze │
└────────────────────────────┼────────────────────────────────────┘
│
┌────────▼─────────┐
│ Netlify Function │
│ bedrock- │
│ analyze.js │
│ │
│ • API key guard │
│ • Daily limit │
│ • JSON validate │
│ • Auto-retry │
└────────┬─────────┘
│
┌──────────────────▼───────────────────┐
│ AWS Bedrock (eu-west-2) │
│ │
│ Amazon Nova Pro (vision + text) │
│ • Step 1: Image → JSON │
│ (defects, cost, risk, citations) │
│ • Step 2: JSON → RICS report │
│ (professional markdown narrative) │
└──────────────────────────────────────┘
| Decision | Rationale |
|---|---|
| Serverless function proxy | API key never exposed to client; enables timeout control and validation layer |
| Sequential image processing | Eliminates React state race conditions that caused crashes with parallel Promise.all |
imagesRef pattern |
Avoids stale closure reads in async callbacks — critical for correct HEIC handling |
| HEIC conversion at upload | Single conversion upstream; both thumbnail and analysis pipeline consume the same JPEG data URL |
| Session fingerprint cache | file.size + hash(first 8KB) + context key prevents re-analysis variance without persisting data |
| JSON schema validation + retry | Malformed AI responses are caught server-side and retried once before surfacing an error |
| Migration to Bedrock | Migrated from Anthropic Claude to AWS Bedrock (Nova Pro) for UK jurisdiction compliance. Same two-step prompt chain, same output quality, compliant infrastructure. |
| Layer | Technology |
|---|---|
| Frontend framework | React 18.3 + TypeScript 5.5 |
| Build tool | Vite 5.4 |
| Styling | Tailwind CSS 3.4 |
| Animation | Framer Motion 12 |
| Icons | Lucide React |
| PDF generation | jsPDF 4.2 |
| HEIC conversion | heic2any 0.0.4 |
| Deployment | Netlify (static + serverless functions) |
| AI Model (Production) | Amazon Nova Pro on AWS Bedrock (eu-west-2) |
| AI Model (Development) | Claude Sonnet 4 (coding assistance only — not in production) |
| API integration | AWS Bedrock Converse API via Netlify function |
- Node.js 18+
- Netlify CLI (
npm install -g netlify-cli) - AWS credentials with Bedrock access (eu-west-2)
# 1. Clone the repository
git clone https://github.com/iamsameershaik/surveyai-taurgo.git
cd surveyai/project
# 2. Install dependencies
npm install
# 3. Create a .env file
cp .env.example .env
# Fill in your BEDROCK_AWS_ACCESS_KEY_ID, BEDROCK_AWS_SECRET_ACCESS_KEY,
# BEDROCK_API_SECRET_KEY, and BEDROCK_DAILY_LIMIT values
# 4. Start the dev server (Netlify CLI required for functions)
netlify devThe app will be available at http://localhost:8888. The Netlify CLI proxies function calls locally so the full analysis pipeline works without deployment.
# Build and deploy to Netlify
netlify deploy --prod --dir=distOr connect the repository to Netlify via the dashboard for automatic deployments on push.
| Variable | Required | Description |
|---|---|---|
BEDROCK_AWS_ACCESS_KEY_ID |
Yes | Your AWS access key for Bedrock. Set in Netlify dashboard under Site Settings → Environment Variables. Never commit this to the repository. |
BEDROCK_AWS_SECRET_ACCESS_KEY |
Yes | Your AWS secret key for Bedrock. Set in Netlify dashboard. Never commit this to the repository. |
BEDROCK_API_SECRET_KEY |
Yes | API key for securing the bedrock-analyze function endpoint against unauthorised calls. |
BEDROCK_DAILY_LIMIT |
No | Maximum number of analyses per day (default: 30). Set higher in production as required. |
These variables are accessed only within netlify/functions/bedrock-analyze.js and netlify/functions/qa.js — they are never sent to or exposed in the browser.
project/
├── netlify/
│ └── functions/
│ ├── bedrock-analyze.js # Main analysis handler — vision → JSON → RICS report (AWS Bedrock)
│ └── qa.js # Report Q&A handler (AWS Bedrock)
├── src/
│ ├── components/
│ │ ├── AnalysisProgressModal.tsx
│ │ ├── ComparisonDashboard.tsx
│ │ ├── DefectHighlightViewer.tsx
│ │ ├── HeroSection.tsx
│ │ ├── ReportQA.tsx
│ │ ├── ReportSection.tsx
│ │ ├── SeverityGauge.tsx
│ │ └── UploadSection.tsx
│ ├── hooks/
│ │ ├── useComparison.ts # Portfolio aggregation + priority ranking
│ │ └── useImageAnalysis.ts # Core analysis pipeline + caching
│ ├── utils/
│ │ └── generatePDF.ts # Full jsPDF report generation
│ ├── types.ts # Shared TypeScript interfaces
│ ├── utils.ts # Prompt builder, compression, API call
│ └── App.tsx
├── netlify.toml
└── package.json
| Criterion | Weight | How SurveyAI addresses it |
|---|---|---|
| AI Integration | 30% | Amazon Nova Pro on AWS Bedrock (eu-west-2) for vision analysis; structured JSON schema with RICS-aligned prompt engineering; two-step prompt chaining (vision → JSON → report); JSON validation with auto-retry; Q&A feature using report context; session-level result caching for consistency |
| Functionality | 25% | End-to-end pipeline from image upload to exportable PDF report; portfolio comparison across multiple properties; HEIC support; defect zone overlay; severity-weighted priority ranking |
| Code Quality | 15% | TypeScript throughout; custom React hooks separating concerns; imagesRef pattern for async state correctness; null-safe rendering; serverless function with validation layer and daily rate limiting |
| UI/UX | 15% | Glassmorphism design system; per-image progress modal; severity gauge; interactive defect overlay with toggle; responsive layout; smooth Framer Motion transitions |
| Documentation | 15% | This README; inline code comments on all non-obvious architectural decisions; reflection document; demo video script |
| Name | Role |
|---|---|
| Sameer Shaik | Lead Engineer — architecture, full-stack implementation, AI integration, deployment |
| Ryan | Documentation — README, submission materials |
| Aditya | Presentation — slide deck, demo narrative |
| Mason | Feature development — Photo Comparison |
| Razeen | QA & Testing — cross-browser testing, bug reproduction |
SurveyAI is designed for triage and prioritisation, not as a replacement for a qualified Chartered Surveyor.
- Every report surfaces an analysis limitations field when image quality is insufficient
- All cost estimates are indicative only and recommend specialist engagement for formal assessment
- RICS standards citations are included in every report to ground outputs in professional frameworks
- No images are stored — files are processed in-session and discarded; nothing is persisted server-side
- The AI is explicitly instructed to decline to fabricate when it cannot make a confident assessment
Built at the Taurgo × Cardiff University AI Hackathon, March 2026.
Image analysis powered by Amazon Nova Pro on AWS Bedrock (eu-west-2, London region) · Deployed on Netlify
Claude by Anthropic used during development for coding assistance and prototyping.
© 2026 Sameer Shaik
This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License.
Commercial use of this software, its source code, or derivative works is not permitted without explicit written permission from the author.
See the LICENSE file for full details: https://creativecommons.org/licenses/by-nc/4.0/