A WhatsApp sales agent for a real-estate company in Mexico. Customers message on WhatsApp, and the agent replies, finds matching properties and books a site visit. I built the n8n workflow, the agent's tools, the API integrations and the deployment.
- The LLM decides when to call each tool: property search, availability check, booking
- Searches live property listings (eBroker API) and free slots in Google Calendar
- Finds the lead in Kommo CRM by phone number, then updates it or creates a new one
- Keeps replies in Spanish or English, remembers the conversation, and handles dates in Mexico's time zone
- Detects when a customer came from a Facebook or Instagram ad
An AI tutor for CBSE Class 8–12 Maths and Science. It talks, writes notes and draws diagrams on an animated whiteboard, asks questions and answers doubts in real time. I built the whole thing myself.
- Uses Gemini, Groq and Claude. If one fails, it switches to the next. Rate limits and daily spend caps keep it running at almost zero cost
- Diagrams come from my own SVG engine (about 16 physics and maths templates), not AI images, so they're always correct
- Numeric and maths answers are checked by a small Python/SymPy service with a PHP fallback, not by asking the AI
- Admin panel for content and analytics, Playwright tests on real lessons, runs on my own VPS with backups
Follows each loan lead from the first enquiry to disbursement: lead capture, document checks, credit processing, approval and payout. Leads are assigned to agents, and admins see how many leads move through each stage.
Covers the full school cycle: admissions, attendance, fees, timetable, exams and results. Admins, teachers and parents each have their own login and see only what they need.
