Smart QS Hackathon 2026 · AI Builders Hackathon

🏗️ Smart QS Copilot

Upload a messy Hong Kong BOQ and follow the money: trade mapping and flags, transparent fair-rate build-ups, a cashflow view, and Security of Payment documents that match the Cap. 652 rules.

Try the live app: smartqs-copilot-app.streamlit.app — upload a messy BOQ and follow it through. Free tier, so it sleeps when idle and can take about 30 seconds to wake. This page is the walkthrough: the demo video, the screens, and the commands to run it yourself in a minute.

Watch the 90-second demo

What it does

1. Upload

Drop in a PDF, CSV or Excel BOQ. It handles the real-world mess: stray title lines, two header rows, merged section labels, item codes, and descriptions that mix English and Traditional Chinese.

2. Trade map and flags

Every line is mapped to a trade and flagged: odd rates, near-duplicate lines, trades with no items, mixed-language descriptions. Each flag cites the item, the number and the reference it deviates from.

3. Fair-rate build-up

Shows how an adjusted fair rate is reached, step by step, from a matched Schedule of Rates item and a project rate index. Nothing is a black box.

4. Cashflow

A payment S-curve with the gaps called out, including the 60-day stretches that put pressure on a subcontractor.

5. Cap. 652 documents

Payment claim and payment response drafted against the Security of Payment Ordinance timelines (30 / 60 / 28 days), plus a pay-when-paid audit.

Screens

Upload screen — Upload a BOQ, or load the sample file that has problems planted in it.

Trade map and flags, with confidence shown per line.

Fair-rate build-up — The fair-rate build-up, step by step.

Cashflow view with the payment gaps called out.

Security of Payment documents — Cap. 652 payment claim and payment response.

Run it yourself

Python 3.10 or newer. Three commands:

git clone https://github.com/BenjaminTia/SmartQS-Copilot
cd SmartQS-Copilot
pip install -r requirements.txt
streamlit run app.py

It opens on http://localhost:8501. No API key is needed: parsing, classification, rate build-ups, flags and dates are all deterministic Python. An optional language model writes one closing paragraph in plain English if a key is present, and the app falls back to a rule-based summary if not.

There is also a test suite (pytest) and an end-to-end smoke check (python smoke_check.py) covering CSV, Excel and PDF inputs.

Honest limits

Screening only. Final pricing and any legal step stay with a qualified quantity surveyor or a lawyer.
Trade mapping follows HKSMM in a loose sense. It is a starting map for a human to correct, not an authority.
Reference rates are a small demo subset. The build-up shows its work so the numbers can be argued with.
It does not claim accuracy and it never hides which rate, rule or clause drove a result.

Links

Source code on GitHub Devpost project page Demo video file