Show HN: Self-hosted DCF workspace using Damodaran datasets, LLM narratives
TL;DR
A developer built a self-hosted stock valuation tool after commercial 'AI analysis' products consistently hid their math or hallucinated inputs.
Key Points
- The tool computes intrinsic value via DCF using Damodaran industry datasets — betas, equity risk premiums, country risk premiums.
- Every assumption is exposed: cost of capital, reinvestment rate, terminal value. LLMs generate bull/bear narratives with cited sources but cannot silently alter the numbers.
- Runs fully local with a single Docker command; code is open source on GitHub.
- Known rough edge: handling high-growth names where DCF structurally breaks down.
Nauti's Take
This is the right way to use AI in financial analysis: a hard separation between what is calculated and what the model merely narrates. Making it structurally impossible for the LLM to silently change the numbers is not a minor detail — it is the most important architectural decision in the whole project.
The DCF breakdown on high-growth names is a real limitation, but that is a fundamental methodology problem that professional analysts routinely paper over too. Using Damodaran datasets as the foundation gives this tool more credibility than most paid products on the market.
Context
Most AI finance tools are black boxes — results look precise but assumptions are hidden or simply invented. This project inverts that principle: deterministic math first, LLM only for context and narrative. That makes errors auditable rather than invisible.
For anyone doing serious valuation work, transparency around terminal value assumptions is not a nice-to-have — it is the core problem.
Sources