Vehify

Note: This project is currently offline due to funding constraints. No live demo is available — the write-up below documents what was built and how it worked.

The Problem

Small and medium businesses lose leads every day because they can’t respond to every DM fast enough. When a customer messages on WhatsApp or Instagram asking about a product, they expect a quick and accurate reply. Hiring staff to manage that around the clock isn’t realistic.

Vehify was built to automate that entire conversation — product discovery, recommendations, and appointment booking — across WhatsApp, Facebook, and Instagram, without any human involvement unless the business wants it.

What We Built

I handled most of the project — both the backend and the React admin panel — with support from the team. The stack is Node.js, Express, and React, with OpenAI powering the AI layer and Meta Business API handling all messaging integrations.

Business onboarding — Companies register on the platform, connect at least one social media account (WhatsApp, Facebook, or Instagram), and set up their product catalog with images, pricing, and properties. They also add company information, a guide on how the AI should interact with customers, and available appointment slots.

AI chat automation — When a customer sends a message to the business’s connected account, the AI responds automatically. It understands context, answers questions, and handles the full conversation without any manual intervention.

Product discovery via tool calling — When a customer asks about products, the AI uses OpenAI tool calling to search the catalog, recommend relevant items, show product images, and explain details — all inside the chat.

Appointment booking — If a customer wants to book an appointment, the AI handles it end to end. It checks availability, collects the customer’s name and contact details if not already known, and confirms the booking automatically.

AI memory — The AI builds a profile for each customer over time — remembering their name, contact details, and previous interactions — so conversations stay personal and context is never lost between sessions.

Admin panel — Businesses can view every conversation in a dashboard. Each chat comes with a sentiment analysis score — positive, negative, or interested — with a confidence rating, so teams can prioritize follow-ups. Admins can also manage and view all appointments.

Technical Challenges

The most complex part was designing the AI tool calling layer — making the AI decide when to search products, when to book an appointment, and when to just respond conversationally, all within a single chat thread. Managing conversation context accurately while keeping responses fast took significant prompt design and iteration.

Integrating with Meta Business API across three platforms — WhatsApp, Facebook, and Instagram — each with their own webhook structure, message formats, and auth flows added real complexity. Every platform needed its own integration while feeding into the same unified AI backend.

Building the AI memory system so the chatbot could remember customer details across sessions and use them naturally — without asking the same questions twice — required careful design to get right.

Results & Impact

Vehify reached a working first version and was built around a real use case — businesses using chat to handle product inquiries and bookings automatically. The project was paused due to funding constraints before going to market. It is currently offline and is a company project with no public source code.