Archived project, Oct 2025 to Nov 2026. Retired November 2026 and replaced by the rebuilt Corporate Research method.
Corporate Research app (v1)
Oct 2025 to Nov 2026
The original SEC research app: AI agents on a cached EDGAR pipeline
The first version of Corporate Research: a web app that pulled SEC filings, ran a team of AI agents over them and produced a research pack with an Excel Monte Carlo DCF.
The question
Can an analyst get from a ticker to a defensible research pack (statements, ratios, risks and a valuation) in minutes rather than days, straight from the primary filings?
The approach
I built a research assistant on SEC EDGAR. It pulled the XBRL data from each 10-K and 10-Q with edgartools, cached it, and passed it to a set of AI agents that wrote up the statements, 18 ratios, the Item 1A risk factors and the segment breakdown. It then produced an Excel DCF with a Monte Carlo layer, so the valuation came out as a range rather than a single number.
Most of the effort went into the data, not the AI. Company filings tag the same item in different ways, carry several years side by side, and report segments that overlap. I wrote a data contract that fixed what each line meant, fallback mappings for when a tag was missing, and checks that removed double counting, for example Amazon's “product” and “service” totals that duplicate its seven revenue streams.
What it showed
The statements came out right. The ratios and risk write-ups were useful first drafts. The lesson came from checking the valuation: when I reviewed the Apple model against the filing, the exporter had taken some balance-sheet lines and the tax figure from the prior year, and equity from the wrong line. Each looked plausible on its own. Together they moved the tax rate from 15.6% to 22.4%.
That is the point of the project. The hard part of financial analysis from filings is not reading them. It is checking that every number belongs to the right period and the right line, and that check is the same whether a person, an app or Claude did the reading.
Retired
The app ran live with no further development from February 2026 until I retired it in November 2026, when Claude could read the filings directly. The method was rebuilt on current tools, with the period checks this version lacked built in from the start. See Corporate Research for the current version.
Sample output
Apple FY2025 research pack, as generated by the original app in early 2026: report, analysis, risk, metrics and statements. Shown as produced, without later review. Superseded by the rebuilt method; the Excel DCF from this version is not shown because some of its inputs came from the wrong year.
Open Source Attribution & Licensing
The research app was built on exceptional open-source projects. We're grateful to the maintainers and contributors of these libraries.
edgartools
Our SEC filing data extraction is powered by edgartools, an outstanding Python library by Dwight Gunning that makes working with SEC EDGAR data elegant and efficient.
MIT License
Copyright (c) 2024 Dwight Gunning
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software.
View Full License →Additional Open Source Libraries
We also rely on these excellent open-source projects:
Research assistance only, not investment advice.
Timeline
- Nov 2026App retired
- Feb 2026Final front-end redesign
- Jan 2026Quarto report package with Python and R Monte Carlo DCF
- Dec 2025Segment data pipeline and cblanalytics.com launched
- Oct 2025Build started on SEC EDGAR with edgartools
Have a problem like this one?
The model is the tool. The analysis is the work. Talk to me about running this kind of analysis on your project.
Get in touch