PREN — Predictive Real Estate Nexus (PREN Lite / Paris Pilot MVP)

PREN is an AI-driven predictive platform that forecasts real estate value and neighborhood gentrification signals up to 5 years ahead using “silent signals” such as zoning/permitting activity and infrastructure plans (with optional privacy-safe aggregated sentiment later).

This repository contains PREN Lite, a Paris pilot MVP built on a serverless AWS stack with:

• a simple scoring API (/score)
• an explainability endpoint (/explain)
• a smoke-test endpoint (/health)
• an ingestion workflow placeholder (Step Functions)
• monitoring (CloudWatch alarm on API 5XX)

Intended use (ethical guardrail): planning & risk management for banks, brokers, and developers.
Not intended for: discriminatory decisions or speculative targeting that could accelerate displacement.

---

AWS 10,000 AIdeas Competition (Building Phase)

Built for the AWS 10,000 AIdeas Competition as part of the Building Phase.

• Team: PREN Systems
• Category: Commercial Solutions
• Pilot city: Paris (Lyon planned as scalability follow-up)

What’s implemented now (PREN Lite):

• Deployed serverless backend (API Gateway + Lambda + DynamoDB + Step Functions + S3)
• Explainable score responses with limitations, ethics, and roadmap
• Health endpoint for end-to-end checks
• CloudWatch alarm for API 5XX errors

Planned next (aligned with the original idea):

• Textract for municipal PDFs → structured signal extraction
• Bedrock (batch) for synthesis/normalization of unstructured “silent signals” (cost-optimized)
• SageMaker (batch) to produce the 5-year Future Value Score + bias audits (Clarify)

---

Live Demo (deployed)

Base URL
https://7nskojt600.execute-api.eu-west-3.amazonaws.com/

1) Score — Future Value Score

Example:

• GET /score?lat=48.8566&lng=2.3522

Returns a demo Future Value Score and IRIS-level pilot fields.

2) Explain — Why the score?

Examples:

• GET /explain?iris_id=PARIS_DEMO_3
• GET /explain?lat=48.8566&lng=2.3522

Returns:

• summary (“why this score”)
• top signals (demo)
• limitations
• ethics & intended use
• next steps + roadmap

3) Health — End-to-end smoke test

Example:

• GET /health

Validates the system can read a known demo score item in DynamoDB.

---

Architecture (PREN Lite)

Region: eu-west-3 (Paris)

Core components:

• API Gateway (HTTP API) routes: /score, /explain, /health
• AWS Lambda (Python 3.11):
  • score_handler.py
  • explain_handler.py
  • health_handler.py
• Amazon DynamoDB
  • Scores table with demo items: PARIS_DEMO_1, PARIS_DEMO_2, PARIS_DEMO_3
  • Signals table reserved for ingestion features
• AWS Step Functions
  • PrenIngestionStateMachine (demo ingestion flow placeholder)
• Amazon S3
  • Raw bucket (PDFs/datasets)
  • Artifacts bucket (batch outputs)
• Amazon CloudWatch
  • Logs (7-day retention)
  • Alarm on HTTP API 5XX errors (prod-minded monitoring)

---

Cost & Safety Notes

• Serverless + DynamoDB PAY_PER_REQUEST keeps costs low for an MVP.
• Log retention is limited (7 days) for cost control.
• No individual-level personal data is processed in this MVP.
• Sentiment is OFF by design for privacy and bias control (RGPD-aware approach later).

---

Roadmap (toward full PREN platform)

• Geo layer: replace demo mapping with real IRIS lookup + optional 500m grid (Paris → Lyon scalability).
• Ingestion pipeline: Textract (PDF) → Bedrock batch structuring into normalized signals (cost-optimized).
• Scoring pipeline: SageMaker batch inference + drift monitoring + bias audit (Clarify) for gentrification risk.
• Optional: privacy-safe aggregated sentiment (opt-in, RGPD-aware) with low weight in the model.

---

Repo Structure

• infra/ — AWS CDK (Python) stack + Lambda handlers
• docs/ — specs / notes used for the competition article

---

License

MIT (add a LICENSE file if you want to make this explicit).