ShipSense AI: Invoice Processing Agent for Enterprise

ShipSense AI: Autonomous Invoice Agent That Validates, Enforces Policies, and Prevents Payment Delays — Automatically

Comment 1

Inspiration

At my employer, an enterprise startup focused on debt financing workflows for Private Equity Firms, Debt Advisors, and Banks, we repeatedly heard a pain point from portfolio companies: manual invoice processing drains hundreds of hours monthly, delaying financial reporting and creating reconciliation headaches. These delays ripple across audits, debt covenants, and liquidity management. Yet enterprises still lose $600B annually to invoice errors and delays. Manual invoice processing is slow, error-prone, and lacks real-time policy checks. We built ShipSense AI to automate this high-velocity, low-criticality workflow — the perfect candidate for AI agent automation.

What it does

ShipSense AI is an autonomous invoice agent that:

  1. Scrapes invoices from emails/PDFs using Apify
  2. Validates amounts, dates, and vendor terms using Gemini's reasoning (stopped mid-day when the network went off, so swapped out with Open AI for some time)
  3. Enforces compliance by cross-checking Senso.ai's policy database (future when we have API access)
  4. Acts proactively (e.g., auto-approves or flags violations, suggests compensation)

hosted onRender Product Design and architecture including System Design and PRD with Open Hands

Demo: [Watch the agent parse an invoice, detect a net-30 policy violation, and generate a discount code in <60 seconds.]

How we built it

  • Scraping & Data Extraction: Apify crawlers for PDFs/emails + custom parsers
  • Policy Engine: Senso.ai Documents API (preloaded vendor contracts)
  • Reasoning & Action: Gemini chain-of-thought prompts to validate fields + suggest fixes
  • Frontend: Streamlit UI showing agent's "thought process" (scraped data → policy checks → action)
  • Mock APIs: Simulated ERP integrations (e.g., NetSuite) for approvals

Challenges we ran into

  • PDF Hell: Handwritten invoices and non-standard formats broke initial Apify scrapers (solved with layout analysis)
  • Policy Ambiguity: Senso.ai required strict schema mapping for Gemini to interpret "net-30" vs "30 days"
  • Over-Automation Fear: Balancing agent autonomy vs human oversight (added "confidence scores" to flag low-certainty cases)

Accomplishments that we're proud of

  • Integrated 3 sponsor tools (Apify, Senso.ai, Gemini) end-to-end in <4 hours
  • Added proactive compensation logic (e.g., "5% discount if payment is late") vs just validation
  • Built a live demo with visible agent reasoning (judges loved the transparency!)

What we learned

  • Real-world data is messy: Even "simple" invoices vary wildly by industry
  • Compliance ≠ Code: Enterprises need audit trails, not just automation
  • UX Matters: Showing the agent’s "thinking" built trust faster than a black box

What's next for ShipSense AI

  • Global Scaling: Support non-English invoices (e.g., EU VAT codes) via Gemini multilingual
  • Legacy ERP Integration: Replace mock APIs with SAP/Oracle connectors
  • Anomaly Detection: Use Rime AI to flag suspicious vendor patterns
  • Multimodal Actions: Send SMS/voice follow-ups via ElevenLabs + Twilio

Built With

Share this project:

Updates