Skip to content

Chunked extraction worker - Extraction Layer #630

Description

@chetanr25

Chunked extraction worker

Parent issue: #550
Blocker issue: #627, #628, #629

The extraction endpoints (#629) queue a job but leave the worker stubbed. This issue builds the real worker: take a queued extraction job and produce a validated contract and a draft incident.

The worker needs to break the input into chunks, run them through the model where a model call is needed and compute the rest directly, validate each result and retry once before marking a chunk failed, then stitch everything into one contract, create the draft incident, and mark the extraction completed. It also has to hold up when a chunk fails outright or Ollama is down.

Tests should run with the model mocked and cover the happy path, a per-chunk retry, and a failure path.

Contract reference

  • contracts/schemas/incident-contract.yaml
  • contracts/schemas/extraction-record.yaml

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    Status
    No status

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions