Day 3 · Indexing

Run the indexing pipeline.

Paste a radiology or imaging report below. The server runs OCR text normalization, extracts the header fields clinicians care about, and tags findings against a curated ICD-10 and SNOMED CT lexicon. Every record is persisted with a deterministic join key that ties it back to its S3 object.

← Back to the roadmap

Try it

Paste report text, then extract and tag.

459 chars · submits to /api/run-index

Results will appear here after the first run.

How it works

  1. 1 · OCR text comes in from a presigned S3 object (when AWS env is set) or from the textarea above. The pipeline is the same either way.
  2. 2 · Normalize line endings, strip control characters, collapse excess whitespace and blank lines so downstream regex can trust the text shape.
  3. 3 · Extract fields from labelled sections ("Patient ID:", "Date of Service:", "Ordering Provider:") with regex, and infer modality and body part from the full text.
  4. 4 · Tag codes against a curated ICD-10 and SNOMED CT lexicon. Each match carries a confidence score that drops when the term is qualified ("rule out", "history of") and that prefers whole-word matches over substrings.
  5. 5 · Persist the index record with a deterministic join key (bucket / s3_key / hash(deterministicId(...))) so documents can be reunited with their S3 object across runs.

Recent runs

What the indexer has produced

0 records

Nothing has been indexed yet. Run the pipeline above to populate this list.