JAVIS Technologies
Multi-Store Data Access Platform
One read-only API for product screens, services, and an AI assistant
9 kinds of data source behind one API
Built with
JavagRPCjOOQMySQLDynamoDBClickHouseAWS
When access cannot be applied safely, the request is refused instead of guessed.
Product screensBackend servicesAI assistant
One read-only API
9 kinds of data source
- The problem
- Product screens, backend services, and an AI assistant all needed the same business data, spread across many databases and services. Each caller wrote its own queries and applied its own company filter, and one missed filter could show one company another company's data.
- My role
- Owned it end to end, from the company's high-level design to working code: the data catalog, the query engine, the data-source connectors, and the access-control layer.
- Approach
- Callers ask for data by name, never by database, and access is checked on every request. The trade-off: when the platform cannot apply someone's access rules safely, it refuses the request instead of guessing.
- Outcome
- One read-only API over nine kinds of data source, for screens, services, and the assistant. No caller writes SQL or filters by company, and every answer comes back complete or with a clear reason why not.
What I built
- One API for every caller. Product screens, backend services, and an AI assistant acting for a user all read data the same way.
- Nine kinds of data source. MySQL, PostgreSQL, CockroachDB, ClickHouse, Trino, DynamoDB, files in S3, HTTP APIs, and gRPC services.
- Joins across databases. One request can combine data that lives in different databases.
- Access control on every request. Each company sees only its own data, sensitive fields can be masked, and nothing can write.
- No silent failures. An answer is complete, or it is refused with a clear reason. It is never quietly cut short.
Questions about this one? Email me, or ask my AI assistant.