Natural language queries over LMS and SIS data
An agentic chat that lets non-technical teams query a multi-tenant LMS and SIS in plain English, with SQL validation before anything runs.
Client
Enterprise EdTech platform
Engagement
Discovery + weekly builds
Tech Stack
Related services
The Challenge
What was the problem?
The useful questions were ordinary: How is enrollment trending? Which courses finish? What does that look like across the SIS and the LMS? The data to answer them sat in a multi-tenant warehouse that spanned both systems. Only engineers could write the joins. Every request from a product, success, or ops teammate turned into a ticket, then a one-off query, then another ticket when the question changed. The constraint was not “add a chatbot.” It was: let non-technical people ask cross-system questions without handing them raw SQL, and without letting a model run unsafe statements against production data.
The Solution
The approach
I built a custom agentic layer that turns a natural-language question into SQL, checks it, and returns a result the asker can read. The agent has a map of the LMS and SIS schema. It can plan a multi-step query, join across those systems, and stop if the generated SQL fails validation. Safety checks sit in front of execution so the chat cannot become a free-form query console. The interface is a conversation, not a report builder. A teammate types the question they would have emailed an engineer. The system answers in seconds when the question is in scope.
Schema map
Documented the multi-tenant LMS and SIS tables the agent is allowed to touch, including how tenants stay isolated.
Agent design
Built a multi-step planner so a question that needs more than one query can still land on a single answer.
SQL generation and checks
Translate English to SQL, then validate before anything runs.
Chat interface
Shipped a simple conversational UI for people who should never see a schema diagram.
Results
What changed
Outcomes from the work as shipped — no extra claims beyond what is on this page.
Non-technical teammates can ask enrollment and completion questions themselves
Cross-system questions that used to wait on an engineer now return in seconds
Fewer ad-hoc data tickets for the engineering team
Self-service analytics across LMS and SIS without a new BI project
Want a working build in week one?
Book a call and I'll help you pick the fastest path to something you can demo.
Ready to build something great?
Every project starts with a conversation. Let's discuss your goals and how I can help you reach them.