United StatesProfessional services2025 – ongoingVerra
AI extraction of legacy carbon-project documents
Verra's document-extraction MVP processed more than 7,000 pages, according to Upstage
Problem
Verra's historical project records were stored in PDFs with inconsistent layouts. Its developers maintained extraction patterns that broke when document formats changed.
Approach
Through the AWS BOX program, Verra worked with Upstage and Pariveda on an extraction pipeline combining OCR and a language model. Developers specified required fields in JSON schemas. The initial MVP processed project descriptions and monitoring reports. After delivery, Verra's developers added another document type themselves.
Results
Takeaway
The initial work reduced dependence on document-specific extraction rules. The reported accuracy covered selected fields in an MVP, rather than the entire historical backlog.
Categories
- Size
- Small business
- Industry
- Professional services
- Business function
- Internal work and knowledge management
- Innovation type
- Process innovation
- AI technology
- Generative AIComputer vision
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Upstage described a delivered extraction MVP developed with Pariveda for Verra. It reported processing more than 7,000 pages and about 90-100% accuracy for critical fields.
Full details are in the original sources.