Live Session: Shipping a RAG Chatbot, Then Training the Team to Run It Alone

Shipping isn't enough

Most contractors deliver a project and then vanish. The result: the client calls back at the first bug, doesn't understand the black box they were handed, and ends up frustrated.

For the Live Session engagement, I flipped the problem around. The deliverable wasn't just a working RAG chatbot — it was a team able to run it, diagnose it and evolve it without me.

The chatbot we shipped

Live Session needed an assistant able to answer questions over a dense, heterogeneous document base. I deployed a custom RAG chatbot with several pieces specific to their use case:

• Smart query reformulation before retrieval, to catch loosely-worded questions
• Automatic OCR of scanned PDFs, to index documents that weren't natively text
• Advanced table handling, often the weak spot of generic RAG pipelines

In production: 95% answer relevance and a response time under 5 seconds.

The real differentiator: training

A high-performing RAG system that nobody internally understands becomes technical debt the day of the first incident. So I structured a knowledge transfer across 4 sessions, from concept to operations.

1. RAG fundamentals

The goal: the team understands the technology instead of just enduring it.

• How embeddings work
• The indexing and retrieval process
• The different types of RAG that exist

2. Their custom implementation

The goal: they fully master their product, not RAG in the abstract.

• Smart query reformulation
• Automatic OCR of PDFs
• Advanced table handling

3. The system's limits

The goal: knowing how to react when things break, instead of panicking.

• The kinds of questions where the system fails
• The documents where RAG hits its limits
• First aid in case of failure

4. Hands-on workshop

The goal: maximum autonomy, under real conditions.

• Database restoration
• Reading and using the Grafana dashboard
• Forecasting and scaling
• The tech watch to keep up

The outcome: autonomy

Today, Live Session no longer needs me for the vast majority of operations. The team masters its product, knows where to look when something goes wrong, and can evolve it on its own.

Thanks to Laurent Janolin and the whole Live Session team for an engagement as technically rich as it was human, and to Ailog for the framework.

About Pierre Kasparian

Pierre is a freelance AI integration and data engineering specialist, and a partner at Ailog. He deploys LLMs, RAG systems and AI agents for SMEs and startups, with one hard rule: never send data outside Europe, in line with GDPR.

To learn more about his work and approach, head to pierrekasparian.com.

---

Want a RAG chatbot your team actually owns? Discover Ailog or talk to Pierre directly.