Description
A course to build assistants that answer questions about company documentation (RAG) and agents with LangChain, using models that run on your own servers. Documents stay in-house and every answer cites its source or admits it does not have one.
Course details
| SIXE code: SXIA05 | Category: Artificial Intelligence / RAG and agents |
| Delivery: Online and on-site | Duration: 2 days |
Who it is for
Developers and architects who need to put an assistant on top of manuals, contracts or internal procedures without sending those documents to an external service.
Prerequisites
Python, REST APIs and JSON. Having used a language model helps, but no previous LangChain experience is needed.
What you will learn
- Preparing the documents: extract text and tables from PDFs with Docling, chunk them and keep track of the page each fragment comes from.
- Retrieving the right content: embeddings, vector and keyword search, and filters so each user only sees what they are allowed to.
- Answering with sources: LangChain chains that cite the fragment used and abstain when there is no evidence.
- Adding tools: a LangGraph agent that queries the index and other APIs, with limits on steps and permissions.
- Measuring: a set of questions with known answers to check whether each change makes the assistant better or worse.
You practise with 25 sample documents, including contradictory versions and unanswerable questions, and finish with an assistant running on a local model.
What you take away
- A document ingestion process with provenance and permissions.
- A RAG assistant built with LangChain and a local model.
- A set of 30 questions to test every change.
If you would rather we build it for you, that is our RAG implementation service with Docling. And if you want to understand why a RAG system ends up answering with stale data, we explain it in this article.
Frequently asked questions
How long is the course?
2 days.
Is it online or on-site?
Both: at your offices or remotely, with a live instructor.
What do I need to know beforehand?
Python, REST APIs and JSON. Having used a language model helps, but no previous LangChain experience is needed.
Does it work with local models or do we need OpenAI?
Both. The course uses a local model and explains what changes when you switch to an external API.
Does RAG eliminate hallucinations?
No. Retrieval, faithfulness to sources and abstention are measured, and acceptance criteria are set from that.
Can it be tailored to my team?
Yes. We adapt the examples and labs to your team’s tools and processes, and it can be combined with other AI courses.
