Skip to content

Reviewer tooling for CDSCO regulatory submissions

सीडीएससीओ नियामक प्रस्तुतियों के लिए समीक्षक उपकरण

Six locked workflows behind a single response envelope. Deterministic rule baselines on every path; LLM polish opt-in and injection-gated. Every metric named in Annexure I §II.5 has a reproducible offline harness in make eval-all.

API reference / एपीआई संदर्भAPI liveschema_version 1.0 · build dev

Submission Guidelines §3.I

Workflows

  1. PS 1

    Anonymise

    POST /v1/anonymise

    Hybrid regex + NLP. HMAC pseudonymisation; optional irreversible generalisation. India-relevant identifiers (Aadhaar, ABHA, PAN, UHID, IMEI).

    KIE micro F1 0.978
  2. PS 2

    Summarise

    POST /v1/summarise

    Three CDSCO source types: SUGAM checklist, SAE narrative, DTAB / DCC meeting minutes. Deterministic extractor; optional LLM polish behind injection gate.

    ROUGE-L 0.253 · BERTScore F1 0.821
  3. PS 3

    Completeness

    POST /v1/completeness

    Locked schema registry. Per-item missing flags + cross-field consistency rules (5 cx-rules: fatal+recovered, non-serious+ICU, etc.).

    accuracy 1.00 · MCC 1.00
  4. PS 4

    Compare

    POST /v1/compare

    difflib + similarity ratio + added/removed lines. Pipe-markdown and TSV tables parsed into row-level table_deltas.

    accuracy 1.00 · MCC 1.00
  5. PS 5

    Classify

    POST /v1/classify

    Severity tiers: death · disability · hospitalisation · other. Routes immediate / priority / standard. SHA-256 + simhash + minhash dedupe.

    Macro-F1 1.00 · MCC 1.00
  6. PS 6

    Inspection report

    POST /v1/inspection-report

    4-section CDSCO template. Numbered observations with severity + recommended actions + due dates. Tesseract OCR for scans.

    OCR CER 0.0058 (Tesseract live)

Annexure I §II.5 - Technical robustness

Live evaluation gates

Reproducible offline. Numbers regenerate on every make eval-all run.

MetricResultSource
k-anonymity / l-diversity / t-closeness1.0 / 5.5 / 0.625last_eval.json
KIE strict micro F1 / macro F10.978 / 0.964last_eval_kie.json
ROUGE-1 / ROUGE-2 / ROUGE-L F10.294 / 0.161 / 0.253last_eval_summarise.json
BERTScore (live)F1 0.821VAIDYAH_BERTSCORE=1
Classification Macro-F1 / MCC1.00 / 1.00last_eval_classify.json
Compare + Completeness MCC1.00 / 1.00respective JSON
OCR CER - synthetic0.0046last_eval_ocr.json
OCR CER - Tesseract live0.0058VAIDYAH_OCR_BACKEND=pytesseract
Segmentation mIoU (synthetic)0.648last_eval_segmentation.json
Latency p95 (8-worker)8.7 – 12 mslast_eval_load.json
Throughput per route875 – 1135 rpslast_eval_load.json
Adversarial robustness33 / 33 across 6 categorieslast_injection_gate.json
PII leak gate0 leaks / 4 docslast_leak_test.json

GET /v1/workflows

Workflow registry