Spaces:

Behrang987
/

myreport-space

Running

App Files Files Community

myreport-space / README.md

Chaudhry Waleed

Deploy RICS v2 (senior baseline, CPU Spaces Dockerfile)

330d570 14 days ago

preview code

Raw

History Blame Contribute Delete

4.22 kB

metadata

title: RICS Report Genius
emoji: 🏠
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: mit
short_description: RICS v2 report generator from past reports + notes.

Report Genius AI

Report Genius AI turns a surveyor's messy site notes (and optional inspection photos) into a structured RICS Home Survey Level 3 report draft — written in the firm's own voice — that the surveyor then reviews, edits, and signs off. It is a Retrieval-Augmented Generation (RAG) system: each firm's own past reports and standard paragraphs shape the wording, while the surveyor's notes remain the only source of facts. Data is isolated per firm ("tenant").

The backend is a Python FastAPI service (backend/, started with uvicorn backend.main:app); the UI is a single static HTML file (frontend/index.html). There is no SQL database — all state is files on disk under DATA_DIR, and the FAISS search index runs inside the API process.

Documentation

Start here: the full technical documentation pack lives in docs/. It takes a new engineer from zero to contributing — architecture, the AI pipeline, retrieval, prompts, APIs, setup, deployment, testing, security, and current status.

Fast links:

New to the project? docs/01 — System overview
Where does code live? docs/02 — Architecture & code map
Getting it running? docs/09 — Development setup
Gentle narrative tour: docs/CODEBASE_GUIDE.md

Quick start

Prerequisites: Python 3.11+, an OpenAI API key (optional for offline dev), and Docker if you want the container path. Full instructions with troubleshooting are in docs/09 — Development setup.

git clone https://github.com/My-Report-AI/Report-genius-ai.git
cd Report-genius-ai
cp .env.example .env          # then set OPENAI_API_KEY and, for local dev, PDF_EXTRACTOR=pypdf

python -m venv .venv
.venv\Scripts\activate        # Windows (macOS/Linux: source .venv/bin/activate)
pip install -e ".[dev]"
pip install -r backend/requirements.txt

uvicorn backend.main:app --reload --port 8000
# open http://localhost:8000  (interactive API docs at /docs)

Docker (local/demo image, bakes the embedding + reranker models for offline start):

docker compose -f docker-compose.v2.yml up --build   # http://localhost:8000

The first local run downloads the local embedding + reranker weights (multi-GB, several GB RAM). See docs/10 — Deployment & configuration for the deploy paths and the full configuration reference.

Repository layout

Path	What it is
`backend/`	The FastAPI backend — all business logic (see docs/02)
`frontend/`	`index.html` (production UI) and `v2.html` (demo UI)
`docs/`	This documentation pack
`scripts/`, `backend/scripts/`	Operator/ingest tooling and dev/deploy helpers
`Dockerfile`	Hugging Face Spaces image (port 7860)
`Dockerfile.v2`, `docker-compose.v2.yml`	Local/demo image (port 8000)
`Master Standard report and paragraphs/`	Operator template bundle (runtime; may be absent locally)

Running tests

python -m pytest backend/tests -q -o addopts=""   # offline; no API key needed

The suite is offline and deterministic by default (LLM/embeddings are stubbed). One caveat: if your local .env sets PII_SCRUBBING_ENABLED=false, a number of PII tests fail by design — see docs/09 and docs/11. Lint/type-check with black, ruff, and mypy backend/.

Deployment

Hugging Face Spaces (current demo): built from Dockerfile, served on port 7860.
Local/demo Docker: Dockerfile.v2 + docker-compose.v2.yml, port 8000.

Details, CI, and environments: docs/10 — Deployment & configuration.

Licence

MIT — see LICENSE.