Finance Automation
Valent
Source-cited preference terms, checked by a validator and confirmed by a reviewer before compute.
About the Project
For cap table and liquidation waterfall work, Valent extracts preference terms from US certificates of incorporation and Indian articles of association, with each value tied to a verbatim quote that a validator checks. A reviewer confirms every term, and the waterfall engine distributes an assumed exit value down the preference stack, with totals reconciled back to the exit value. The output distributes an assumed exit value. It does not estimate fair market value, which needs an EV-to-equity bridge and a locked valuation date.
Technical Overview
Data Sources
- Certificates of incorporation (DOCX)
- Articles of association (text PDF, XFA forms, scanned PDFs)
- Cap table spreadsheets (DOCX)
- Embedded eAOA attachments
ML / NLP Stack
- LLM structured extraction (gpt-4o-mini via OpenRouter, Ollama fallback)
- Verbatim citation grounding
- Clause indexing and keyword clause selection
- OCR (EasyOCR service)
- Brute-force convert-or-hold equilibrium
Key Challenges
- Seniority and pari-passu ranking, which current models get wrong
- Defined-term chains across Indian articles
- OCR noise breaking verbatim citations
- Lakh and paise number formats
Interested in this solution?
Let's discuss how we can adapt this approach for your business.
Get in Touch