An AI assistant grounded in the course videos

I built an AI assistant for The Vault so students could ask questions and get answers from the course videos, with links to the relevant video and timestamp. I was the sole developer across transcription, search, answer generation, conversation memory and student context.

My role: Sole developer, from pipeline to product
Built for: Students inside The Vault
Areas of work: Python, RAG search, source citations, memory & student context
Support: Maintenance following the main development phase

Course knowledge, through conversation

01
Ask
02
Find
03
Explore

An answer with a path back to the lesson.

Students can ask a question, get an answer from the course material and follow a video timestamp to learn more.

01 / The starting point

Make hours of course content easier to navigate.

The course videos contained the information students wanted, but finding a particular answer could mean searching through hours of content. I built the agent to give students a more direct way into that knowledge.

Inside The Vault, students could ask questions, get answers drawn from the course material and follow the source video and timestamp to explore the lesson in more depth.

The Vault course assistant answering a question about the Direction First framework — Course questions and answers

02 / What I built

The full system behind the conversation.

01Transcription & search

Turn videos into searchable knowledge

I built the video transcription pipeline and split transcripts into searchable sections. This gave the assistant a way to find relevant parts of the course material in response to a student's question.

02RAG answers

Build answers around the course material

I built the retrieval and answer-generation workflow around the course content. The aim was to let students ask a question and get an answer drawn from the videos, making the existing library easier to use.

03Video citations

Keep the original lesson within reach

I linked answers back to the relevant video and timestamp. Students could check the source, watch the explanation in context and continue learning from that point in the course.

04Memory & profiles

Give the conversation context

I added conversation memory and student-profile context so the assistant could follow an ongoing discussion and tailor its responses to the student's learning context.

05Python service

Own the service and its maintenance

I built the agent as a separate Python service for The Vault and was its only developer. Development ran from late 2024 through 2025, followed by maintenance of the project.

Python development
RAG workflows
Product integration
Maintenance

The Vault course assistant's answer with a course-part reference and a named document listed under Sources — Course source references

A follow-up conversation in The Vault course assistant — Follow-up conversation

03 / The result

A question could lead straight to a lesson.

Students gained a way to ask questions across the course library, with answers linked to the source material. Conversation memory and student context helped tailor the experience to their learning.

Feedback on the assistant was positive. After the main development work from late 2024 through 2025, my work on the project continued through maintenance.

Have an AI feature to build?

Let's work out what comes next.

Discuss a project Back to projects