GATE — The Last Gate Before Payment

Built for small engineering startups: one local AI tool to cross-check invoices, orders & project docs before you pay—no AP department, no cloud upload. You decide; GATE does the reading.

Comment

Inspiration

Most small engineering startups do not have a procurement department, a compliance team, or even a full-time AP clerk.

On a team of five or fifteen, the founder, project lead, or office manager is often the same person who must verify contractor invoices, check equipment orders, and make sure files match company policies and regulations—on top of running the business.

That work is repetitive and easy to get wrong under pressure. One mismatched PO number, W-9 typo, or missing sign-off can delay payment, strain subcontractor relationships, or block project cash flow.

Generic AI can summarize a PDF, but it will not know your rules—and many startups cannot send project financials to cloud AI at all.

We built GATE for teams like this: a local, one-person-friendly compliance gate before payment. AI handles the tedious cross-checking; you keep the final approve/hold decision.

What it does

GATE is a local-first review tool for small engineering startups and SME firms—not an enterprise ERP replacement.

Built for lean teams

  • No IT department required: ./gate.sh start and you are running
  • Onboarding in minutes: enter company name, load a few policy PDFs, start uploading
  • One person can run the full workflow from upload → review → email vendor → approve payment

What you upload

  • Contractor invoices, equipment orders, W-9s, change orders, progress billing
  • Excel schedules, Word letters, or a ZIP of everything a subcontractor sends at once

What GATE does (so you do not have to read every line)

  • Reads each file and compares numbers and references across documents
  • Flags compliant / warning / critical items against rules you ingested (LightRAG graph—not generic cloud prompts)
  • Shows where issues are on PDFs when possible
  • For ZIP packages: cross-document checks (e.g., EIN on W-9 vs. invoice, PO qty vs. packing list) + Micro Report chart
  • Decision Trace: plain-language steps explaining why something is held—not a black box
  • Human-in-the-loop: check the findings you agree with, draft a vendor email, you click approve

The last gate before payment Startups should not pay until documents align with policy—but they also cannot afford weeks of manual review. GATE helps a small team catch errors faster, decide with more confidence, and pay vendors without hiring a compliance department.

Our demo also contrasts GATE with vanilla RAG and generic local chat: graph-grounded, cross-file, traceable—not “summarize this PDF and hope.”

How we built it

We optimized for small teams with limited time and zero DevOps:

Piece Choice
Run locally Ollama + FastAPI—no SaaS subscription, no document egress
One command gate.sh starts models, backend, and auto-restart supervisor
Knowledge LightRAG ingests your PDF policies—no bundled fake “demo violations”
Files PyMuPDF + native Excel/Word parsers; ZIP pipeline in gate_package.py
UI Single-page app—upload, review, copilot, no training manual required

Stack: Python, FastAPI, LightRAG, Ollama (Llama 3.1, mxbai-embed-large), PyMuPDF, JavaScript.

Challenges we ran into

  • No dedicated users: UX had to work for a non-technical project manager, not a procurement specialist.
  • No policy team: GATE only helps after you upload a few real standards—but we made ingest async with clear progress so one person can kick it off and keep working.
  • Cross-file pain at SME scale: Subcontractors send ZIPs, not neat single PDFs; package integrity had to be first-class.
  • Trust on a lean team: The person who signs the check needs to see why—Decision Trace and manual confirmation were mandatory.

Accomplishments that we're proud of

  • A startup-sized workflow: one person can set up, audit, and decide
  • Live multi-format + ZIP audit with cross-document matching
  • 100% on-device— viable for bootstrapped firms that refuse cloud document upload
  • Open source: github.com/Shao116/GATE

What we learned

Small engineering startups do not fail reviews because they lack intelligence—they lack time and specialized headcount. Accessible AI means a three-person firm can get the same kind of pre-payment scrutiny large contractors pay departments for, as long as the tool stays local, grounded in their policies, and respectful of human final authority.

What's next for GATE

  • Starter policy packs for small construction/engineering firms (W-9, change orders, retainage basics)
  • Simpler “first-week setup” wizard for teams with no compliance background
  • Vendor resubmission checklist (plain English) so subcontractors fix issues in one round
  • Lightweight shared workspace when a startup grows from 1 to 3 people touching AP

Built With

Share this project:

Updates