Aviral GargLocal prototypes · Synthetic data · October 2026
CareerPartner
Career context that keeps corrections connected to their sources.
The prototypes explore how to build current career context while retaining the records behind it.
This is a project overview. The context-packet command runs offline on synthetic data; a separate work-story demo runs locally. A public interactive demo is not available here.
Read how context is built ↓The offline
context packet
- Read the fixture
Start from a read-only synthetic JSON fixture with attributed source records.
- Resolve corrections
Build deterministic current context, retaining the original records.
- Keep unknowns visible
Export corrected current claims alongside source attribution and explicit unknowns.
A work story,
reviewed by its author
After an explicit Finish story, the local demo can request one optional, source-bound Muse clarification. Offline examples are also available. The person can edit the account, review it and explicitly accept it into local notes.
A model question is a proposal, not an accepted fact. Actual microphone capture and audible playback still need a human check.
Before export
The prototype rejects duplicate identifiers, dangling correction targets, cycles and competing correction branches. Ambiguous corrections stop the export instead of silently choosing a claim.
Hosted product features are still under consideration. CareerPartner remains a prototype.