feat: add Document Intelligence workflow

document-ai/document-intelligence/.gitignore

@@ -0,0 +1,2 @@
+outputs/
+

document-ai/document-intelligence/README.md

@@ -0,0 +1,10 @@
+# Document Intelligence
+
+An offline document-processing workload over a small synthetic procurement
+dataset. It ingests semi-structured text invoices, classifies them, extracts
+fields, reconciles totals, detects duplicate invoice numbers, and emits an
+auditable JSONL/CSV report. It uses only Python's standard library.
+
+Run `./run.sh`. The files in `data/documents/` are the versioned test dataset.
+Replace that directory with a larger local corpus to create a long-running run.
+

document-ai/document-intelligence/data/documents/invoice-001.txt

@@ -0,0 +1,9 @@
+DOCUMENT: INVOICE
+INVOICE_ID: INV-2026-001
+SUPPLIER: Atlas Scientific Ltd
+DATE: 2026-09-14
+ITEM: Sensor array | 3 | 420.00
+ITEM: Shielded cable | 10 | 18.50
+TAX: 216.75
+TOTAL: 1661.75
+

document-ai/document-intelligence/data/documents/invoice-002.txt

@@ -0,0 +1,9 @@
+DOCUMENT: INVOICE
+INVOICE_ID: INV-2026-002
+SUPPLIER: Meridian Compute SA
+DATE: 2026-09-18
+ITEM: GPU workstation | 2 | 2450.00
+ITEM: NVMe storage | 4 | 210.00
+TAX: 861.00
+TOTAL: 6601.00
+

document-ai/document-intelligence/data/documents/purchase-order-001.txt

@@ -0,0 +1,6 @@
+DOCUMENT: PURCHASE_ORDER
+ORDER_ID: PO-2026-031
+SUPPLIER: Atlas Scientific Ltd
+DATE: 2026-09-10
+APPROVED_LIMIT: 1800.00
+

document-ai/document-intelligence/docker/Dockerfile

@@ -0,0 +1,6 @@
+FROM python:3.12-slim
+WORKDIR /app
+COPY data /app/data
+COPY scripts/pipeline.py /app/scripts/pipeline.py
+ENTRYPOINT ["python", "/app/scripts/pipeline.py"]
+

document-ai/document-intelligence/environment.yaml

@@ -0,0 +1,6 @@
+id: document-intelligence-environment
+name: Document Intelligence local environment
+driver: local
+runtimes: [{id: local, type: local, containerRuntime: docker}]
+resources: [{id: workstation, runtimeId: local, cpuCores: 4, memoryBytes: 4294967296}]
+

document-ai/document-intelligence/run.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+root=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd); rm -rf "$root/outputs"
+python3 "$root/scripts/pipeline.py" --output "$root/outputs" --stage all
+"$root/validate.sh"
+

document-ai/document-intelligence/scope.yaml

@@ -0,0 +1,4 @@
+id: document-intelligence-scope
+name: Document Intelligence local scope
+environmentVersionIds: [document-intelligence-v1]
+

document-ai/document-intelligence/scripts/pipeline.py

@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+import argparse, csv, hashlib, json
+from pathlib import Path
+ROOT=Path(__file__).resolve().parents[1]
+def dump(p,v): p.parent.mkdir(parents=True,exist_ok=True); p.write_text(json.dumps(v,indent=2)+"\n")
+def ingest(out):
+ docs=[]
+ for p in sorted((ROOT/'data/documents').glob('*.txt')):
+  text=p.read_text(); docs.append({'file':p.name,'sha256':hashlib.sha256(text.encode()).hexdigest(),'text':text})
+ dump(out/'ingested.json',docs)
+def classify(out):
+ docs=json.loads((out/'ingested.json').read_text())
+ dump(out/'classified.json',[{**d,'type':next(x.split(':',1)[1].strip().lower() for x in d['text'].splitlines() if x.startswith('DOCUMENT:'))} for d in docs])
+def extract(out):
+ result=[]
+ for d in json.loads((out/'classified.json').read_text()):
+  fields={}; items=[]
+  for line in d['text'].splitlines():
+   if ': ' not in line: continue
+   key,val=line.split(': ',1)
+   if key=='ITEM':
+    name,qty,price=[x.strip() for x in val.split('|')]; items.append({'name':name,'quantity':int(qty),'unit_price':float(price)})
+   else: fields[key.lower()]=val
+  result.append({'file':d['file'],'sha256':d['sha256'],'type':d['type'],'fields':fields,'items':items})
+ dump(out/'extracted.json',result)
+def reconcile(out):
+ docs=json.loads((out/'extracted.json').read_text()); rows=[]; seen=set()
+ for d in docs:
+  if d['type']!='invoice': continue
+  f=d['fields']; subtotal=sum(i['quantity']*i['unit_price'] for i in d['items']); total=float(f['total']); tax=float(f['tax']); iid=f['invoice_id']
+  rows.append({'invoice_id':iid,'supplier':f['supplier'],'subtotal':subtotal,'tax':tax,'total':total,'balanced':abs(subtotal+tax-total)<.01,'duplicate':iid in seen}); seen.add(iid)
+ dump(out/'reconciliation.json',rows)
+ with (out/'invoice-summary.csv').open('w',newline='') as fp:
+  w=csv.DictWriter(fp,fieldnames=list(rows[0])); w.writeheader(); w.writerows(rows)
+def report(out):
+ rows=json.loads((out/'reconciliation.json').read_text()); ok=all(r['balanced'] and not r['duplicate'] for r in rows)
+ dump(out/'report.json',{'status':'passed' if ok else 'failed','documents':len(json.loads((out/'ingested.json').read_text())),'invoices':len(rows),'gross_total':sum(r['total'] for r in rows)})
+ artifacts=['ingested.json','classified.json','extracted.json','reconciliation.json','invoice-summary.csv','report.json']
+ dump(out/'manifest.json',{'showcase':'document-intelligence','status':'success' if ok else 'failed','artifacts':artifacts})
+STAGES={'ingest':ingest,'classify':classify,'extract':extract,'reconcile':reconcile,'report':report}
+def main():
+ p=argparse.ArgumentParser();p.add_argument('--stage',choices=[*STAGES,'all'],default='all');p.add_argument('--output',required=True);a=p.parse_args();o=Path(a.output);o.mkdir(parents=True,exist_ok=True)
+ for s in list(STAGES) if a.stage=='all' else [a.stage]: STAGES[s](o)
+if __name__=='__main__':main()
+

document-ai/document-intelligence/topology.yaml

@@ -0,0 +1,3 @@
+id: document-intelligence-topology
+name: Document Intelligence local topology
+links: []

document-ai/document-intelligence/validate.sh

@@ -0,0 +1,11 @@
+#!/bin/sh
+set -eu
+root=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)
+python3 - "$root" <<'PY'
+import json,sys
+from pathlib import Path
+r=Path(sys.argv[1]);m=json.loads((r/'outputs/manifest.json').read_text());q=json.loads((r/'outputs/report.json').read_text())
+assert m['status']=='success' and q['documents']==3 and q['invoices']==2 and q['gross_total']==8262.75
+assert all((r/'outputs'/x).is_file() for x in m['artifacts']);print('validated document-intelligence')
+PY
+

document-ai/document-intelligence/workflow.yaml

@@ -0,0 +1,11 @@
+name: document-intelligence
+spec:
+  namespace: showcase
+  image: ghcr.io/uffescience/akoflow-examples/document-intelligence:v1
+  activities:
+    - {name: ingest, runtime: local, command: {entrypoint: python, arguments: ["/app/scripts/pipeline.py", "--stage", ingest, "--output", "."], workingDirectory: /workspace}, cpuLimit: "1", memoryLimit: 256Mi}
+    - {name: classify, runtime: local, command: {entrypoint: python, arguments: ["/app/scripts/pipeline.py", "--stage", classify, "--output", "."], workingDirectory: /workspace}, cpuLimit: "1", memoryLimit: 256Mi, dependsOn: [ingest]}
+    - {name: extract, runtime: local, command: {entrypoint: python, arguments: ["/app/scripts/pipeline.py", "--stage", extract, "--output", "."], workingDirectory: /workspace}, cpuLimit: "2", memoryLimit: 512Mi, dependsOn: [classify]}
+    - {name: reconcile, runtime: local, command: {entrypoint: python, arguments: ["/app/scripts/pipeline.py", "--stage", reconcile, "--output", "."], workingDirectory: /workspace}, cpuLimit: "1", memoryLimit: 256Mi, dependsOn: [extract]}
+    - {name: report, runtime: local, command: {entrypoint: python, arguments: ["/app/scripts/pipeline.py", "--stage", report, "--output", "."], workingDirectory: /workspace}, cpuLimit: "1", memoryLimit: 256Mi, dependsOn: [reconcile]}
+