Key Takeaway
Chrono built an AI leadership coach that answers from the coaching company's internal knowledge base through RAG and tailors each answer to the manager asking, and program adoption rose by 60%.
A leadership coaching company had a method that worked in the room. Between sessions, managers drifted. Chrono built an AI coach that answers from the company’s own knowledge base and tailors every answer to the manager asking. Program adoption went up 60%. If your expertise lives in documents and people’s heads, this is how you put it in front of every user, every day.
Why wasn’t the program getting used between sessions?
Because nothing was there when managers needed it. A hard conversation with a direct report doesn’t wait for the next session, and a manager with a real problem won’t dig through course material to find the answer.
A generic chatbot wasn’t the fix either. It gives generic advice, and generic advice contradicts what the program teaches. The company needed a coach that sounded like its own coaches.
How does the AI coach use the company’s knowledge base?
Through retrieval-augmented generation, or RAG. Before the model answers, the coach searches the company’s internal knowledge base, pulls the passages that fit the question, and answers from those. The program’s method is the source, not the open internet.
The knowledge base is indexed in Pinecone, a vector database that finds passages by meaning rather than exact keywords. GPT-4o writes the reply from what comes back. Every conversation is traced in LangSmith, so the team can see which passages the coach used and check that its advice matched the method.
How does the coach tailor answers to each manager?
It starts from the manager, not from the content. Two managers asking about the same problem get different answers, because the coach takes their own situation and the conversation so far into account before it pulls from the knowledge base. The AI service also writes a summary of each coaching session, so the thread carries over.
The result feels like a coach who knows the program and knows you. That’s what kept managers coming back.
What changed for the company?
Program adoption rose 60%. Managers used the program between sessions, not only during them, and the company’s method reached people at the moment they needed it.
What is the architecture?
The AI coach runs as its own Python service, separate from the main backend, so the team can change prompts, models or retrieval without touching the app. Redis and BullMQ job queues move slow work off the request path and keep the app responsive when traffic spikes.
What else did it take to ship?
A full product around the coach. Chrono built six apps in one monorepo: the bilingual mobile app, a REST API, a headless CMS so the team updates content without developers, a payments admin, the AI service and an internal Slack bot. It shipped on iOS and Android in French and English, and Chrono now runs it through managed app support.
What does this mean for your company?
If your edge is what your team knows, an AI coach or assistant grounded in that knowledge is the fastest way to scale it. The model is the easy part. The work is in the knowledge base, the retrieval and the tracing that proves the answers are right.
See how we build these systems on our retrieval and knowledge page. To scope your own, start here.
Frequently Asked Questions
How did an AI coach increase program adoption by 60%?
It made the program available between sessions, when managers actually face problems. The coach answers from the company’s own method and tailors each answer to the manager, so the advice is useful enough to come back for.
What is RAG on an internal knowledge base?
Retrieval-augmented generation means the AI searches your own documents before answering and replies from what it finds. Here the coach searches the coaching company’s knowledge base, so it gives the company’s advice, not the internet’s.
How do you check that an AI coach gives the right advice?
Trace every conversation. On this platform, LangSmith records which passages the coach retrieved and how it used them, so the team can review answers against the program’s method.
