Domain-agnostic AI analytics platform
Five layers from ingestion to output, with domain adapters for healthcare, elections, FMCG and surveys, and forecasting and anomaly detection built in.
The problem
Healthcare, elections, FMCG and survey analytics need the same pipeline: get the data in, model it, reason about it, present it. What differs is the domain knowledge.
What we built
A platform in five layers, where the domain is a plug-in rather than a rewrite:
- Ingestion: bring data in from the sources each domain uses.
- Data: a DuckDB analytical store.
- Local ML: forecasting and anomaly detection that run on the platform itself.
- LLM reasoning: an LLM layer that reasons over the results.
- Output: dashboards in Superset.
Domain adapters for healthcare, elections, FMCG and surveys supply the vocabulary, metrics and data mappings for each field.
Architecture
Each layer has one job and a clear contract with the next, so a layer can be replaced without touching the others. Adding a domain means writing an adapter.