Learner onboarding chatbot

Client: Tech training institution, East Africa
My role: Tech lead and automations developer
Stack: FastAPI Google Cloud Laravel nginx OpenAI pgvector PostgreSQL Redis

This project was delivered for a confidential client. Names, data and identifying screens have been removed or recreated.

The context

New learners join through an online student portal. In their first weeks, they need answers about schedules, tools, requirements and where to find things.

The problem

The same onboarding questions arrived again and again, and support staff answered them one by one. Learners waited; staff lost time they needed for harder cases.

What we built

A FastAPI service that answers questions using OpenAI's gpt-4o-mini, grounded in the institution's own content. Answers are retrieved from a PostgreSQL 16 database with pgvector, and Redis caches frequent requests. The service runs on Google Cloud and is embedded directly in the existing Laravel/Livewire portal, so learners never leave the page they're on.

I set up the database server and placed pgAdmin behind nginx, so the team can administer the data without exposing the database to the internet.

How we shipped it

The chatbot went live through a structured rollout: a pilot with formal sign-off, a go-live plan, and a business continuity plan with a RACI matrix that names who does what if something fails. A master document index ties it all together.