Vasool AI

Inspiration

We kept watching freelancer friends lose evenings chasing "bhai payment kar dena" on WhatsApp. Grey's research says 8 in 10 freelancers face delays, and Zerodha's Daily Brief puts MSME receivables in lakhs of crores with almost no tool for the solo operator. We built Vasool so the awkward chase runs itself while you do billable work.

What it does

Vasool AI is an autonomous collections agent for Indian micro-businesses and freelancers. You add overdue invoices, activate the agent, and it follows up on WhatsApp in Hindi, Hinglish, Tamil, or English: negotiating dates, dropping UPI links, and logging every decision. You pay 5% only on money actually recovered.

How we built it

We shipped a Next.js 14 dashboard with Tailwind and shadcn/ui, a file-backed ledger for invoices and immutable agent logs, and API routes that call Gemini 2.5 Flash for tone and negotiation decisions. WhatsApp Cloud API sends live messages when keys are present; Razorpay links and webhooks close the loop. Everything runs in Docker for Cloud Run deploy.

Challenges

WhatsApp template compliance forced us to separate demo threads from live sends while keeping one code path. Vernacular escalation was trickier than English dunning: we encoded politeness curves per days-overdue and language so the agent stays firm without sounding rude.

Accomplishments

End-to-end demo where a judge can activate an agent, reply as a debtor in Hinglish, get a commitment plus UPI link, simulate payment, and watch recovered totals and reasoning logs update with zero manual steps.

What we learned

Collections AI is really a scheduling and culture problem wrapped in LLM text. Function-calling style JSON from Gemini made actions (commitment, payment link, mark paid) auditable, which matters more for XPRIZE judging than prettier chat bubbles.

What's next

Invoice forwarding and OCR ingestion, voice escalation via WhatsApp calling, TReDS-style batching for micro-vendors, Hindi-first mobile onboarding, and Razorpay Route for automatic success-fee splits.

Built With

• cursor
• firestore/postgres
• gemini-2.5-flash
• gemini-api
• google-cloud-run
• next.js
• python/node-backend
• razorpay/upi
• tailwind-css
• whatsapp-cloud-api