Publish artifacts for multimodal-document-retrieval-20260812

README.md

@@ -0,0 +1,59 @@
+---
+license: cc-by-4.0
+language:
+- en
+pretty_name: Multimodal Document Retrieval Baseline Synthetic Evaluation Set
+size_categories:
+- n<1K
+task_categories:
+- visual-document-retrieval
+tags:
+- synthetic
+- multimodal-ai
+- evaluation
+- visual-document-retrieval
+- document-question-answering
+- image-to-text
+- feature-extraction
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train.jsonl
+  - split: test
+    path: data/test.jsonl
+---
+
+# Multimodal Document Retrieval Baseline Synthetic Dataset
+
+## Summary
+
+This dataset contains 14 training examples and 4
+held-out examples for **Business documents contain meaning in text, tables, layout, and imagery that text-only retrieval can miss.**
+
+Every record is synthetic and includes:
+
+- `input`: query, event, or feature description
+- `label`: expected class, route, relation, or evidence category
+- `context`: synthetic supporting context
+- `source`: fictional source identifier
+- `variant`: generation pattern
+- `synthetic`: always `true`
+
+## Uses
+
+- Reproducible unit and integration tests
+- Baseline model training
+- Evaluation harness development
+- Schema and architecture demonstrations
+
+## Limitations
+
+The starter dataset contains synthetic textual modality descriptors, not sensitive scanned documents.
+
+This dataset does not represent real users, patients, customers, production
+traffic, or licensed media. It must not be presented as real-world evidence.
+
+## Related Model
+
+[RKB109/multimodal-document-retrieval-20260812-model](https://huggingface.co/RKB109/multimodal-document-retrieval-20260812-model)

data/test.jsonl

@@ -0,0 +1,4 @@
+{"id":"multimodal-document-retrieval-2-1","input":"Locate the architecture diagram showing a queue","label":"technical-diagram","context":"Caption service diagram; labels producer queue consumer.","source":"doc-02","variant":"direct","synthetic":true}
+{"id":"multimodal-document-retrieval-3-2","input":"In an operations review, Find the signed privacy exception request","label":"policy-form","context":"OCR privacy exception; layout signature block approved.","source":"doc-03","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-4-3","input":"For an evaluation case, Which invoice contains a tax adjustment?","label":"invoice","context":"OCR invoice 184; table subtotal tax adjustment final total.","source":"doc-04","variant":"evaluation","synthetic":true}
+{"id":"multimodal-document-retrieval-6-1","input":"Which form includes retention approval?","label":"policy-form","context":"OCR retention request; checkbox approved; signature present.","source":"doc-06","variant":"direct","synthetic":true}

data/train.jsonl

@@ -0,0 +1,14 @@
+{"id":"multimodal-document-retrieval-1-1","input":"Find the invoice with a shipping surcharge table","label":"invoice","context":"OCR total 1480; table shipping surcharge 80; image company seal.","source":"doc-01","variant":"direct","synthetic":true}
+{"id":"multimodal-document-retrieval-1-2","input":"In an operations review, Find the invoice with a shipping surcharge table","label":"invoice","context":"OCR total 1480; table shipping surcharge 80; image company seal.","source":"doc-01","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-1-3","input":"For an evaluation case, Find the invoice with a shipping surcharge table","label":"invoice","context":"OCR total 1480; table shipping surcharge 80; image company seal.","source":"doc-01","variant":"evaluation","synthetic":true}
+{"id":"multimodal-document-retrieval-2-2","input":"In an operations review, Locate the architecture diagram showing a queue","label":"technical-diagram","context":"Caption service diagram; labels producer queue consumer.","source":"doc-02","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-2-3","input":"For an evaluation case, Locate the architecture diagram showing a queue","label":"technical-diagram","context":"Caption service diagram; labels producer queue consumer.","source":"doc-02","variant":"evaluation","synthetic":true}
+{"id":"multimodal-document-retrieval-3-1","input":"Find the signed privacy exception request","label":"policy-form","context":"OCR privacy exception; layout signature block approved.","source":"doc-03","variant":"direct","synthetic":true}
+{"id":"multimodal-document-retrieval-3-3","input":"For an evaluation case, Find the signed privacy exception request","label":"policy-form","context":"OCR privacy exception; layout signature block approved.","source":"doc-03","variant":"evaluation","synthetic":true}
+{"id":"multimodal-document-retrieval-4-1","input":"Which invoice contains a tax adjustment?","label":"invoice","context":"OCR invoice 184; table subtotal tax adjustment final total.","source":"doc-04","variant":"direct","synthetic":true}
+{"id":"multimodal-document-retrieval-4-2","input":"In an operations review, Which invoice contains a tax adjustment?","label":"invoice","context":"OCR invoice 184; table subtotal tax adjustment final total.","source":"doc-04","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-5-1","input":"Show the network diagram with a gateway","label":"technical-diagram","context":"Caption network topology; labels client gateway service.","source":"doc-05","variant":"direct","synthetic":true}
+{"id":"multimodal-document-retrieval-5-2","input":"In an operations review, Show the network diagram with a gateway","label":"technical-diagram","context":"Caption network topology; labels client gateway service.","source":"doc-05","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-5-3","input":"For an evaluation case, Show the network diagram with a gateway","label":"technical-diagram","context":"Caption network topology; labels client gateway service.","source":"doc-05","variant":"evaluation","synthetic":true}
+{"id":"multimodal-document-retrieval-6-2","input":"In an operations review, Which form includes retention approval?","label":"policy-form","context":"OCR retention request; checkbox approved; signature present.","source":"doc-06","variant":"operations","synthetic":true}
+{"id":"multimodal-document-retrieval-6-3","input":"For an evaluation case, Which form includes retention approval?","label":"policy-form","context":"OCR retention request; checkbox approved; signature present.","source":"doc-06","variant":"evaluation","synthetic":true}

project.json

@@ -0,0 +1,52 @@
+{
+  "name": "Multimodal Document Retrieval Baseline",
+  "problem": "Business documents contain meaning in text, tables, layout, and imagery that text-only retrieval can miss.",
+  "domain": "multimodal-ai",
+  "architecture": "retrieval",
+  "hugging_face_tasks": [
+    "visual-document-retrieval",
+    "document-question-answering",
+    "image-to-text",
+    "feature-extraction"
+  ],
+  "recommended_stack": [
+    "FastAPI for ingestion and multimodal search",
+    "Transformers with LayoutLMv3 or ColPali-style encoders",
+    "OCR adapter plus image captioning",
+    "PostgreSQL plus pgvector for fused embeddings",
+    "Object storage for source documents",
+    "OpenTelemetry plus retrieval ablation reports"
+  ],
+  "real_world_data_sources": [
+    {
+      "name": "SEC EDGAR company facts API",
+      "url": "https://data.sec.gov/api/xbrl/companyfacts/CIK0000320193.json",
+      "purpose": "Real tabular filing facts and document metadata"
+    },
+    {
+      "name": "arXiv API",
+      "url": "https://export.arxiv.org/api/query?search_query=all:document%20understanding&start=0&max_results=5",
+      "purpose": "Public papers and document-layout retrieval scenarios"
+    }
+  ],
+  "job_description_skills": [
+    "Multimodal embeddings and document understanding",
+    "OCR, layout, image, and text feature fusion",
+    "Vector retrieval and modality-aware ranking",
+    "Ablation testing and retrieval evaluation",
+    "Scalable document ingestion and object storage"
+  ],
+  "impact_targets": [
+    "Improve Recall@5 by >= 20% over text-only retrieval",
+    "Report modality ablations for OCR, layout, and image signals",
+    "Index 100,000 documents with resumable ingestion",
+    "Keep p95 retrieval latency below 500 ms"
+  ],
+  "baseline_evaluation": {
+    "test_examples": 4,
+    "accuracy": 1,
+    "synthetic_evaluation": true
+  },
+  "estimated_delivery": "8-12 weeks for one engineer",
+  "generated_baseline_is_production_ready": false
+}

sources.json

@@ -0,0 +1,12 @@
+[
+  {
+    "name": "SEC EDGAR company facts API",
+    "url": "https://data.sec.gov/api/xbrl/companyfacts/CIK0000320193.json",
+    "purpose": "Real tabular filing facts and document metadata"
+  },
+  {
+    "name": "arXiv API",
+    "url": "https://export.arxiv.org/api/query?search_query=all:document%20understanding&start=0&max_results=5",
+    "purpose": "Public papers and document-layout retrieval scenarios"
+  }
+]