Chatterbox: B2B Language Learning Platform

Chatterbox connects refugees and migrants who are native speakers of less-commonly-taught languages with corporate learners. The founders, Mursal and Guillemette, had the concept validated and partnerships lined up. There was an early prototype, but it was basic and development had stalled. The previous engineering setup wasn't moving at the speed the business needed.

I came in and took over the technical side. The job was to rebuild and scale the platform into something that could actually support the business: tutor scheduling, video sessions, contract management, seat licensing, learner progress tracking. All of it needed to work and it needed to work fast.

The core platform is Django and React. Tutors set their availability, learners book sessions, and everything runs through Twilio for video and chat. Stripe handles billing. Google Calendar syncs appointments. The database has over 100 tables covering users, organizations, contracts, courses, assessments, and ratings.

As the client base grew from individual learners to enterprise contracts (Microsoft, PwC, VISA, Unilever, Red Cross, Georgetown University), I built a separate manager dashboard. Corporate managers needed to allocate seat licenses, track engagement across their organization, and export reports. That dashboard runs on TypeScript, React, and TanStack Table, backed by a serverless API on AWS Lambda with Prisma.

The data side got its own system too. I built an ETL pipeline in Python that pulls from the production database, transforms it, and loads it into a separate data warehouse. Student lifecycle tracking, tutor quality metrics, license utilization forecasting, invitation funnel analysis. All of it runs on scheduled Lambda jobs.

Two things that came later and show how the role evolved from "build the product" to "build the platform."

First: a natural language database query service. Non-technical staff needed answers from a 100+ table database and couldn't write SQL. I built a FastAPI service that uses LangChain and GPT-4 to figure out which tables are relevant, generate the query, and recover from errors automatically. If the first query hits a missing column, it re-fetches the actual schema and tries again.

Second: an image generation service for course content. Language experts needed images for lessons and were spending hours on stock photo sites. I built a service that takes a prompt, hits four providers in parallel (DALL-E 3, Stability AI, Pexels, Unsplash), and streams results back in real-time over WebSocket. The language experts just type "French restaurant dining scene" and pick from a dozen options.

The company started as B2C. Individual learners buying sessions. It wasn't scaling. The decision to pivot to B2B (selling seat licenses to enterprises) changed everything about the product: pricing, onboarding, user management, reporting, the concept of "an organization" that manages learners under a contract.

I led the technical side of that pivot. Rebuilt the billing model, added the contract and licensing system, built the manager dashboard from scratch. At the same time, the founder was raising a seed round, which meant investor conversations about architecture, scalability, data security, and technical roadmaps. I was the person answering those questions.

The round closed. The company grew to 60+ enterprise clients across education, financial services, consulting, and NGOs. Over 6,000 learners have used the platform.

I'm still advising as fractional CTO. The platform runs, the enterprise clients are active, and the team has grown. What started as taking over a stalled prototype turned into four years of product evolution: real-time video, enterprise dashboards, AI-powered tools, and a data warehouse.