MoneyForge
All Posts
Digital Products

How to Build a Document Intelligence Service With doc7

MoneyForge Team 2026-08-14 9 min read

Traditional document converters extract text. They see a chart and output gibberish. They see a formula and lose the structure. They see a diagram and skip it entirely. This is why RAG systems fed with converted documents give terrible answers — the conversion loses 40-70% of the information.

doc7 launched in August 2026 and hit 1,100+ stars quickly. Built by magicrew, it takes a fundamentally different approach: instead of extracting characters, it uses a multimodal vision model to understand each page as a whole — reading charts, formulas, diagrams, tables, and even handwritten annotations, then outputting structured Markdown that preserves the meaning.

In a benchmark on two raster-only PDFs with 15 checkable visual facts, doc7 recovered 15 out of 15 facts. The same test with MarkItDown (Microsoft's document converter with OCR) scored 9 out of 15. Docling scored 3 out of 15. The difference is visual understanding versus text extraction.

What doc7 Does Differently

Featuredoc7Traditional OCR (Tesseract, ABBYY)Text Extraction (anydoc, MarkItDown)
TextYesYesYes
TablesPreserved structureOften brokenPreserved
Charts and graphsUnderstands and describesSkippedSkipped
Mathematical formulasConverts to LaTeXSkipped or garbledSkipped
Diagrams and flowchartsDescribes relationshipsSkippedSkipped
Scanned documentsYes (visual understanding)Yes (but error-prone)Limited
ScreenshotsYesLimitedNo
Handwritten notesPartial (model-dependent)PoorNo
Requires OCR stackNoYesNo
Uses your own modelYes (OpenAI-compatible)N/AN/A

The key advantage: doc7 runs with your own local or remote multimodal model. You can use a free local model through Ollama or LM Studio, which means zero per-page cost. There is no per-page document-processing fee from doc7 itself.

Business Model: Academic and Research Document Processing

Researchers and students deal with PDFs constantly — papers, textbooks, theses, lab reports. These documents are full of charts, formulas, and diagrams that traditional tools cannot parse. Universities and research labs need searchable, AI-ready versions.

ServiceWhat You DeliverPriceTarget
Single paper conversionOne PDF to structured Markdown$5-15 per paperIndividual researchers
Literature review processing50-100 papers converted and indexed$200-800PhD students, research teams
Textbook digitizationFull textbook to searchable Markdown$100-500 per bookUniversities, online courses
Conference proceedings200+ papers batch processed$500-2,000Conference organizers

Your cost: $0 if using a local model. Your time: roughly 1-2 minutes per page (mostly automated). A 20-page paper takes 20-40 minutes to process. At $10 per paper, that is $15-30/hour.

Law firms have filing cabinets full of contracts, court filings, and case documents. These need to be searchable, comparable, and analyzable. Traditional digitization loses the structure of legal tables, redlines, and embedded exhibits.

ServiceWhat You DeliverPriceTarget
Contract digitizationContract PDFs to structured Markdown$50-200 per contractLaw firms, in-house legal
Case file processingFull case file organized and searchable$500-2,000 per caseLitigation teams
Due diligence document review100+ contracts analyzed for key terms$1,000-5,000M&A teams, VCs
Compliance document managementAnnual conversion + indexing$200-800/month retainerCompliance departments

Legal documents are high-value (a single contract may be worth millions), so clients pay premium rates for accurate digitization. The key selling point: doc7 preserves tables, exhibits, and cross-references that text extraction loses.

Business Model: RAG Pipeline Builder

Every company wants an AI chatbot that knows their documents. The pipeline requires: document ingestion, conversion to clean text, embedding, and retrieval. doc7 handles the conversion step better than any other tool because it preserves visual information.

Project ScopeDocumentsYour PriceDelivery Time
Small (50-200 documents)Internal policies, FAQs, manuals$1,000-3,0003-5 days
Medium (500-2,000 documents)Research papers, reports, archives$3,000-8,0001-2 weeks
Large (5,000+ documents)Full knowledge base migration$8,000-25,0002-4 weeks
Ongoing maintenanceMonthly new documents$500-2,000/monthContinuous

Your pipeline: doc7 converts documents to Markdown, OpenAI embeddings create vector representations, Pinecone or Weaviate stores them, and a chatbot retrieves and answers. The quality of the RAG system depends heavily on the conversion step — this is where you add value.

Business Model: Medical Record Processing

Healthcare providers need to digitize patient records, lab reports, and medical imaging reports. These documents contain charts, tables, and diagrams that are critical to patient care. Traditional OCR misses or mangles these elements.

Document TypeVolumeYour PriceNotes
Patient intake forms100-500 per month$200-600/monthHIPAA compliance required
Lab reports50-200 per month$300-800/monthCharts and reference ranges
Medical imaging reports50-100 per month$400-1,000/monthDiagrams and measurements
Clinical trial dataProject-based$2,000-10,000/projectHigh accuracy required

Medical document processing requires HIPAA compliance (in the US) or equivalent data protection regulations. Use doc7 with a local model so no data leaves the client's premises. This is a major selling point over cloud-based services.

Income Projections

These estimates assume you focus on one primary vertical and work 25-35 hours per week.

PhaseRevenue SourceMonthly Income
Month 1-2First conversion projects$500-2,000
Month 3-4RAG pipeline projects + retainers$2,000-6,000
Month 5-6Specialized vertical (legal or medical)$4,000-12,000
Month 7+Multiple verticals + enterprise contracts$8,000-25,000

What You Need to Get Started

RequirementCostNotes
doc7Free (MIT license)The conversion tool
Multimodal modelFree (local via Ollama/LM Studio) or $0.01-0.05/page via APIqwen3.5-9b works well locally
Computer with GPU (optional)Your existing laptop may workLocal models run on CPU, GPU is faster
No cloud requiredEverything runs locally

Total startup cost: $0 if using a free local model. This is one of the lowest-cost businesses to start.

Common Mistakes

Promising perfect accuracy. doc7 is significantly better than alternatives but not flawless. Complex diagrams with overlapping elements, very low-quality scans, and unusual formatting can still cause errors. Always offer a human review step for critical documents.

Competing on volume instead of quality. Do not try to be the cheapest document converter. You are selling visual understanding — the ability to preserve charts, formulas, and diagrams that competitors lose. Charge for that value, not per page.

Ignoring vertical specialization. General document conversion is a commodity. Legal document processing is worth 5-10x more per page because the stakes are higher. Medical document processing commands premium prices because of compliance requirements. Pick a vertical and become the expert.

Not demonstrating the difference. Show potential clients a side-by-side comparison: the same document processed by traditional OCR versus doc7. The visual difference is striking — tables preserved, charts described, formulas intact. This demonstration closes deals.

The document intelligence market is projected to reach $15 billion by 2028. The shift from text extraction to visual understanding is just beginning. doc7 gives you a tool that outperforms enterprise solutions costing $10,000+/month — for free, running on your own hardware. The businesses built around it now will define the standard for document intelligence services.