The problem we solve
Every day, technology integrators and IT companies in Latin America lose hours — and clients — because of slow quotations.
A potential client sends a WhatsApp message: "I need a firewall for 30 users and a PBX for 10 extensions." What happens next? The sales person has to stop what they're doing, look up product specs, check pricing lists (in USD and ARS separately), calculate margins, write the document, format it, and send it. That process takes 45 to 90 minutes — and often happens the next day.
The client already called three competitors in that time.
What GMT QuoteAgent does
GMT QuoteAgent is a multi-agent AI system that compresses that 90-minute workflow into under 5 minutes — automatically, 24/7, without human intervention.
A client sends a message via WhatsApp or email. Three AI agents work in sequence:
- Agent 1 — Classifier (Gemini Flash): Reads the raw message, extracts intent, product category, quantities, and client context. Returns structured JSON.
- Agent 2 — Quote Builder (Gemini Pro + Claude API): Queries the real GMT product catalog, selects the right products, calculates pricing in both ARS and USD with live exchange rate, and drafts a professional quotation document.
- Agent 3 — Follow-up (Gemini Flash): Monitors the CRM daily. If a sent quotation receives no response after 3 days, it generates and sends a personalized follow-up automatically.
Every operation is timestamped and logged. The client receives a branded PDF. GMT gets an entry in the CRM with full context.
Why this is different
Most quoting tools are generic SaaS built for US or European markets. GMT QuoteAgent is built from 20+ years of real B2B technology sales experience in Argentina:
- Native ARS/USD dual pricing with configurable daily exchange rate
- Real catalog data: Fortinet, Grandstream, Yeastar, Dahua — the actual products technology integrators sell
- Argentine commercial context: IVA handling, payment terms, quotation validity windows
- WhatsApp-first: the channel Latin American SMB clients actually use, not email forms
This is not a demo built for a hackathon. GMT Telecomunicaciones is the first production user. Every quotation GMT generates during the hackathon period runs through this system.
How we built it
The orchestration layer runs on N8N (self-hosted via Docker), which handles webhook ingestion, agent chaining, and CRM writes. The product catalog lives in NocoDB alongside the CRM and the full operations evidence log.
The AI pipeline:
- Incoming WhatsApp message → Twilio webhook → N8N
- N8N sends message to Gemini Flash (Google Cloud) for classification
- Classified JSON → Gemini Pro for catalog matching and pricing
- Draft text → Claude API for document quality control and formatting
- HTML quotation → PDF generation → delivery via WhatsApp API + email
- All metadata written to NocoDB with full timestamp
The Gemini API is the core of the system. It handles both the intent classification (Flash model for speed and cost) and the catalog-aware quotation generation (Pro model for reasoning quality).
Challenges we faced
The ARS/USD problem: Argentine pricing is inherently dual-currency. Products are purchased in USD from distributors and invoiced in ARS to clients. Building a quotation engine that handles this correctly — with configurable exchange rate, margin by category, and IVA — required careful prompt engineering to avoid calculation errors.
WhatsApp message variability: Clients don't write structured requests. "Necesito algo para la red de mi empresa, somos 20 personas" (I need something for my company network, we're 20 people) is a real input that the classifier must handle. Training the Gemini classifier on real GMT inquiry history was essential.
Evidence architecture: XPRIZE requires proof of real business operation. We designed the NocoDB schema from day one to log every agent call, response time, client interaction, and commercial outcome — not as an afterthought.
What we learned
Building an AI-native business is fundamentally different from adding AI to an existing business. Every workflow had to be redesigned from the input channel up. The N8N + NocoDB + Gemini stack proved fast to iterate — we went from zero to first real quotation in under two weeks.
The biggest insight: domain knowledge is the moat. Any developer can call the Gemini API. Only someone with 20 years of technology sales experience in Argentina can build a quotation agent that actually generates accurate, professional, commercially viable quotes for Latin American clients.
What's next
GMT QuoteAgent is the MVP of a larger vision: an AI-native business operating system for technology SMBs. The quotation workflow is the entry point. Next phases include AI-driven CRM follow-up, licitaciones (government tender) analysis, and supplier price monitoring — all running as agents in production.
Log in or sign up for Devpost to join the conversation.