AutopilotProcure

Inspiration

SMEs constantly lose resources (money, labor,..) to manual procurement and hidden price hikes. We built AutopilotProcure to give every business an autonomous AI agent that defends their margins with hands-free sourcing. MNCs also have problem to consider the best choice of bidding option from suppliers.

What it does

It instantly extracts data from messy PDF quotes, benchmarks prices across the hidden web, sends proactive price-increase alerts, and physically drafts orders automatically on local supplier portals.

How we built it

Orchestration & AI: OpenAI, Dify, and Manus Data Extraction (OCR): Interfaze (Jigsawstack) and Fractal Scraping & Benchmarking: Exa and BrightData Browser Automation: TinyFish and ByteRover Backend & DB: Google Cloud UI & Design: Lovable, Sleek.design, and Yboard Dev Stack: GitHub Copilot, and Antigravity Pro, Codex Desktop app (all using GPT Codex)

Challenges we ran into

Extracting structured data from highly irregular PDF invoices required heavy iteration with specialized OCR. Additionally, orchestrating long-running browser agents (TinyFish) required us to build a robust asynchronous polling architecture to prevent backend timeouts.

Accomplishments that we're proud of

Closing the "automation loop." We didn't just build an advisory dashboard—we built a true agent that reads a quote, finds a cheaper alternative, alerts the user, and autonomously drafts the final purchase order.

What we learned

We learned that chaining together specialized micro-models (deep-web search, deterministic OCR, browser agents) is exponentially more powerful and reliable for B2B tasks than relying on a single massive LLM.

What's next for AutopilotProcure

Integrating directly with ERPs like SAP, Oracle or other ERPs to auto-reconcile orders. We also plan to train predictive models on global supply chain news to warn buyers of price hikes before they happen.

Built With

• codex
• githubcopilot
• javascript
• next.js
• openai
• tailwindcss
• tinyfish
• typescript