Inspiration

A business can make sales and still struggle to pay its bills when customer payments arrive late. We wanted to address this practical challenge with AI that supports action—not just conversation.

Our idea is CashFlow Sentinel, an agentic AI assistant designed to help small businesses understand outstanding invoices, identify payment disputes, and prepare evidence-based follow-ups while keeping people in control.

What it does

CashFlow Sentinel is designed to turn invoice records, payment ledgers, customer emails, and collection policies into a prioritized receivables action plan.

The planned prototype will:

Calculate outstanding balances and overdue days.
Identify partially paid, overdue, disputed, and promised-to-pay invoices.
Connect customer communications to the relevant invoices.
Recommend the next action with supporting evidence.
Draft customer-specific reminders or dispute-clarification messages.
Require human approval before an action is finalized.
Display a rule-based estimate of expected collections, with assumptions clearly shown.
Maintain an audit trail of recommendations, evidence, and approval decisions.

How we plan to build it

We plan to use Python and Streamlit for the application, SQLite for structured records, and a free-tier language-model API for email understanding and draft generation.

The workflow will separate financial calculations from language-model reasoning. Python functions will calculate invoice balances, aging, and priority scores. The language model will interpret customer messages and draft responses using the supplied evidence.

The agent will follow a constrained workflow:

Upload records → validate data → calculate receivables → interpret customer messages → retrieve evidence → recommend an action → request human approval.

To keep the 24-hour build realistic, the MVP will focus on CSV invoices, ledger records, pasted customer messages, and a short collection policy. PDF extraction will be a stretch feature.

Challenges we expect to address

Our main challenge is making recommendations trustworthy. A reminder should not ignore a recorded payment or an unresolved dispute.

We plan to address this through structured data validation, evidence-linked recommendations, deterministic financial calculations, and a “Needs human review” status when information is missing or contradictory.

Another challenge is avoiding unsafe autonomy. The prototype will generate drafts rather than automatically sending emails, changing invoices, or making financial commitments.

What makes it different

CashFlow Sentinel is not intended to be another document chatbot. Its focus is a complete, constrained business workflow: reconcile records, prioritize cases, explain recommendations, prepare follow-ups, and obtain approval.

Its value will be demonstrated through traceable decisions and useful actions—not simply fluent AI responses.

What we aim to demonstrate

Using synthetic business records, our demo will show:

An overdue invoice that needs a reminder.
A partially paid invoice requiring a corrected balance.
A disputed invoice requiring clarification instead of escalation.
A promised payment that should be monitored.
Human review and approval of the recommended follow-up.

We will evaluate balance-calculation correctness, invoice-status classification, evidence coverage, and whether actions follow the supplied collection policy.

What’s next

Future development could include accounting-system integrations, multilingual drafts, improved document extraction, and collection estimates validated against historical payment data.

The long-term goal is to make receivables management more understandable, actionable, and accessible to small businesses.

Our Vision: Help every SME make smarter financial decisions, recover payments faster, and maintain healthier cash flow.