EdTech / aviation
An AI-powered learning platform for aviation training
We paused the BI work we were paid for to migrate the platform first — then built the KPI warehouse and shipped Eddy, an agentic assistant deployed as a first-class Open edX XBlock.
Client
A specialised aviation training software provider
5–6×
increase in company valuation
The challenge
What was in the way
A leader in flight-simulation software was running its learning platform on WordPress. It could not collect granular learning data, could not analyse student performance, and could not scale.
The solution
What we built
We told them not to. Building analytics on a platform that could not emit the data was spending money twice, so we paused the BI work mid-stream and led a migration to Open edX first — a decision that cost us billable weeks and saved the client a rebuild.
On that foundation we built the whole ecosystem: Open edX on Azure Kubernetes Service with an environment-separated release pipeline, an xAPI ingestion layer feeding a ClickHouse KPI warehouse with automatic type inference and ingest-log lineage, generating seven tiers of learning KPIs, and Superset on top for instructors.
Then we productised the AI. Eddy is an agentic assistant shipped as a first-class Open edX XBlock and Tutor plugin — one-command deployable, not a bolted-on widget. It plans in chain-of-thought, generates SQL, and queries the warehouse through an MCP toolbox, so instructors ask questions in English and get answers from live learning data. It runs on OpenAI or on local Ollama, so institutions whose data cannot leave the estate can still use it.
We also built the mission delivery layer: a custom XBlock with JWT-signed mission tokens and dual-cloud object storage.
The assistant is now on its second generation, and the reason we could replace the first one safely is that its behaviour was never a matter of opinion: every run is traced and logged, so a change in answer quality shows up as evidence rather than as a complaint.
Key technologies
- Platform
- Open edX, Tutor, Azure Kubernetes Service, Azure Container Registry
- Data
- Python, ClickHouse, PostgreSQL, dbt, Celery, Superset
- AI
- MCP Toolbox text-to-SQL, OpenAI and local Ollama
- Observability
- LangSmith tracing, structured run logs, pinned model contracts
- Content
- xAPI/Tin-Can, JWT, CLA/JWKS, S3/MinIO + Azure Blob
Impact & results
Company valuation up 5–6×
On the client’s own assessment — repositioned from a service provider into a scalable EdTech platform.
A production AI assistant with real data access
Not a chatbot bolted to a FAQ.
Learning analytics that did not previously exist
Seven KPI tiers from event data the old platform could not emit.
Cloud-native and auto-scaling
Deployed through a guarded release pipeline.
Start here
Start with a problem, not a brief.
Tell us what is slow, what is manual, what is stalled, or what nobody understands any more. We will tell you honestly whether it is worth building.