techem-horizon

<div align="center"> <img width="320" alt="techem-horizon" src="https://github.com/user-attachments/assets/56ffc738-94df-4e6c-8f9d-7e11941a4426" /> </div> # Techem Horizon > Portfolio-intelligence platform for real-estate energy data. <img width="1710" height="898" alt="techem2" src="https://github.com/user-attachments/assets/8f8247d1-76b2-4460-b745-30473895a468" /> <br /> <img width="1709" height="895" alt="techem1" src="https://github.com/user-attachments/assets/95850f00-406c-49ff-a68c-99ad9fb198cd" /> <br /> Raw meter readings get stitched together with live weather, geometry and geo data to produce a living view of every building and apartment — with interactive overviews![Uploading techem1.png…]() , forecasts **Techem MCP** as the AI chatbot on top. Three surfaces: - **Backend** — FastAPI service that owns all numbers (KPIs, histories, weather-driven forecasts). - **Frontend** — React + Vite app: 3D map, isometric buildings, floor plans, charts. - **Techem MCP** — natural-language chatbot powered by OpenAI function-calling over the backend's tools. ## The idea Techem sits on millions of building readings. Horizon shows what happens when that data meets **public context** (Open-Meteo weather, OSM geometry, Nominatim geocoding) and lands in a product a portfolio manager actually wants to open. Three bets: 1. **Weather is the story.** Residential demand follows outside temperature. Every forecast fits `kWh = k·HDD + base` on real Open-Meteo data. 2. **Gaps are not blockers.** Missing readings? Deterministic HDD-based synthesis keeps the UX consistent. 3. **One source of truth.** Backend owns numbers. Frontend is a lens. MCP is a second lens, asked in English. ## Architecture ```mermaid flowchart LR subgraph External OM[Open-Meteo] OSM[OpenStreetMap] MT[MapTiler] OAI[OpenAI] end SB[(Supabase<br/>properties · units · rooms · readings)] BE[FastAPI backend<br/>HDD × linear forecast] FE[React frontend] MCP[Techem MCP] SB --> BE --> FE OM --> BE OSM --> BE MT --> FE SB --> FE BE <--> MCP OAI <--> MCP MCP --> FE ``` ## Features - **Portfolio map** — 3D MapLibre with OSM footprints, heat-colored extrusions, click-to-fly. - **KPI cards** — annual kWh, €, CO₂ per building, color-coded vs. portfolio average. - **Building detail** — isometric building (click per apartment), 2D floor plan, room-level pie chart, comparison + forecast charts at monthly / weekly / daily. - **Weather-driven forecast** — actual-vs-forecast timeline overlaid with building average and temperature. - **Techem MCP chatbot** — OpenAI function-calling over live backend tools (portfolio stats, rankings, anomaly scans, full report). ## Repo ``` backend/ FastAPI + Supabase + Techem MCP frontend/ React + Vite app ``` - [backend/](backend/README.md) — data model, HDD forecast, APIs - [frontend/](frontend/README.md) — pages, map, charts - [Techem MCP](backend/app/mcp/README.md) — OpenAI agent, tools, chat protocol ## Quick start ```bash # backend cd backend && python -m venv ../.venv ../.venv/bin/python -m pip install -r requirements.txt ../.venv/bin/uvicorn app.main:app --reload --port 8000 # frontend cd frontend && npm install && npm run dev ``` Open `http://localhost:5173`. Env, schema bootstrap, and deployment live in the sub-READMEs. ## Stack React 19 · Vite · Tailwind · shadcn/ui · Recharts · deck.gl · MapLibre · FastAPI · Pydantic · Supabase Postgres · Open-Meteo · Nominatim · OpenAI · Railway · Vercel. ## Scope This project was created during the [Futury Build Days 2026](https://www.starthub-hessen.de/de/events/futury-build-days/). Challenge given by [Techem](https://www.techem.com/).