AVAI
How It Works
Upload a filing
Submit a company's financial filing as a PDF (preferred) or an Excel file. AVAI reads it and extracts the key figures for checking.
AI & Python Verify
AI reads the filing and extracts the figures. Python code then checks every figure against the filing before anything is built on it.
Synthesis
Multiple AI reviewers challenge the case, resulting in a final synthesis that drafts the investment memo.
Inside the council
1
Reviewers
Separate reviewers build the case for the company, the case against it, and a sceptical view that tests the assumptions. Where they disagree, the memo shows it.
2
The Outside Angle
A separate role brings a perspective from beyond the filing. Its findings are labelled as outside research and kept apart from what the filing proves.
3
The Integrity Check
Looks at management and governance concerns, and at what the filing does not disclose. Where there is nothing to assess, it says so instead of guessing.
4
The Auditor
Tests every claim against the filing and marks it supported, unsupported or contradicted. Weak findings are flagged in the memo, not hidden.
5
The Python Layer
Code, not AI, checks each extracted figure against the filing and runs the valuation with stated assumptions.
6
The Synthesis
A final step reads everything, disagreements included, and writes the investment memo: recommendation, confidence level, valuation range, conditions and missing evidence.
AI reads, Python verifies.
Investment Memo Structure
Each memo is structured into clear sections that explain the case, the numbers behind it, and what to watch next.
Recommendation
- A clear stance on the investment, with an explicit confidence level and the conditions attached to it.
Valuation
- A cautious, central and optimistic valuation range, with the key assumptions behind each scenario shown.
Strengths & Weaknesses
- Strengths and weaknesses drawn mainly from figures verified against the original filing, with any outside research labelled separately.
Risks & Opportunities
- Potential growth and hazards, based on the report data and an external view maintained as distinct.
Outside angle
- Context from beyond the filing, kept separate from what the filing proves.
Conditions and monitoring
- What must be confirmed before committing more, and what to track next.
About the Project
Built by Priyadarshan Sengottaiyan. I built AVAI to shorten the first pass of company research. Analysts spend days pulling figures from filings, checking them and looking at a company from several angles. AVAI does that first pass in minutes and gives the analyst a structured memo to challenge, not a verdict to accept.
The principle behind it is 'AI reads, Python verifies.' AI is good at reading long filings and code is good at checking, so every figure is matched against the filing by code before anything is built on it, and every claim is marked supported, unsupported or contradicted.
AVAI is designed for private equity and venture capital due-diligence workflows. Today it is a working prototype, demonstrated on public company filings, and I improve it with every real filing I run. It supports an analyst's judgement and does not replace it. It is not investment advice.
Demo Access Request
Request a one-time access code to explore the prototype.
Prototype notice: filings are read in memory and never saved, but the demo keeps extracted figures and memos until the server restarts. Filings go to AI models under zero data retention. Please upload public filings only.