About the role
You help a person find and use their own information without losing its source, its meaning, or its privacy. The measure is whether the answer is useful and correctly attributed, which is a different and harder target than embedding similarity.
The work
Build ingestion, parsing, retrieval, structured memory and provenance across documents and selected services. Preserve permissions at retrieval time and distinguish original records from generated summaries. Handle corrections, conflicting facts and deletion across indexes and caches.
What good looks like
In your first 90 days, deliver a personal search workflow with source-linked answers, permission checks and a verified deletion path.
Evidence we look for
Bring information-retrieval, database or ML application experience. Show how you measure answer usefulness and source correctness rather than only embedding similarity.
What we need to see
- Information retrieval, database, or applied machine-learning experience
- You measure answer usefulness and source correctness rather than only similarity
- You keep provenance attached through the whole pipeline
- You can evaluate retrieval quality on a corpus that belongs to one person
Nice to have
- Hybrid retrieval combining lexical and vector search
- Personal knowledge management or note systems
- On-device indexing under storage limits
The exercise
Resolve a question when two documents disagree and only one may be shared with the requesting agent.
Where and how we work
In the office together five days a week, in any of these cities. Remote-friendly around your family, arranged one person at a time.