hef.aws.monce.ai

HEF Extraction Pipeline

AI-powered document extraction for surface treatment. Delivery notes, purchase orders, inspection reports — any format, any language. Powered by Claude VLM + Snake SAT matching.

▶ Open Upload UI ■ Dashboard 📃 API Docs
-
Workers
-
Queued
-
Completed
-
Failed
📄

Multi-Format Ingestion

PDF, images (PNG/JPG/TIFF), Excel, Word, Outlook MSG/EML. Drop any mix of files — they all get converted and extracted.

🌐

Multilingual VLM

French, German, English, Italian, Swedish, Dutch — Claude reads them all. Handwritten annotations and stamps included.

🔍

Snake SAT Matching

16,522 customers + 20,119 articles. Supplier identification via name, logo, SIRET. Part numbers matched to catalog in <5ms.

Surface Treatment Focus

DLC, PVD, HSG, nitriding, metallisation — treatment types auto-detected from descriptions. Built for HEF/Techniques Surfaces.

🚀

Async Pipeline

4 parallel workers. Submit via POST, poll via GET. Multi-file batches auto-synthesize. Long-poll support for instant results.

Smart Routing

AUTO_APPROVE for clean extractions, HUMAN_REVIEW for flagged anomalies. Confidence scores on every field and match.

6-Stage Pipeline

Stage 0
Supplier ID
Haiku + Snake
Stage 1
Classify
Haiku VLM
Stage 2
Extract
Sonnet VLM
Stage 3
Normalize
Rules
Stage 4
Match
Snake SAT
Stage 5
Validate
Rules + Flags

Quick Start

# Single file extraction
curl -X POST https://hef.aws.monce.ai/extract \
  -F "file=@delivery_note.pdf"
# {"task_id": "abc-123", "status": "queued"}

# Multi-file batch
curl -X POST https://hef.aws.monce.ai/extract \
  -F "files=@note1.pdf" -F "files=@order.xlsx"

# Get result (long-polls until done)
curl https://hef.aws.monce.ai/extract/{task_id}