/ project · interactive demo
stock-vetter
An LLM-driven research engine for fundamental stock analysis.
ProblemDoing real fundamental analysis on a company means reading hundreds of pages of SEC filings and earnings transcripts. This is slow work that's easy to do inconsistently. I wanted a tool that could perform a rigorous, repeatable first pass: pull the primary sources, reason about them the way a value investor actually does, and surface only what deserves a human's attention.
What it doesFor any ticker it fetches the company's SEC filings and analyst-call transcripts, runs a structured multi-pass LLM analysis scoring the business across six dimensions, cross-checks the numbers with a reverse-DCF to expose the growth the market is implying, and produces a weighted verdict with its reasoning shown rather than hidden.
What I learnedThe hardest failures lived in the data, not the model. Careful review of outputs surfaced a fiscal-year bug that corrupted results for companies that don't report on a December calendar, and a parser quietly feeding the wrong 10-Q sections into the model. Fixing those, plus prompt caching and adaptive sampling that cut cost per ticker by roughly 40%, drove home that in LLM pipelines correctness lives in the source-data plumbing.
How a ticker gets vetted
Captured example. Not live output, not financial advice. This is a frozen snapshot of one real NVDA run, shown to illustrate the tool's output and methodology. The numbers are fixed captured values, not live data, and nothing here is a recommendation to buy or sell any security.
Step 1Six-dimension score
Moat
7.5/10why 7.5CUDA's installed base of 7.5M developers and full-stack co-design create real lock-in, but hyperscaler custom ASICs and China foreclosure keep it contestable.
what to look forHow protected the profits are from competition. The wider the moat, the more durable future cash flows.Owner-earnings
7.5/10why 7.5FCF tracks net income closely and converts well, but a $8.9B non-cash equity gain and fast-growing stock-based comp distort reported earnings upward.
what to look forNet income plus non-cash charges, minus the maintenance capex the business genuinely needs. A cleaner read on what an owner could pocket than reported EPS.Capital allocation
5.5/10why 5.5Buybacks and R&D are funded entirely from free cash flow, but $17.5B into illiquid startups and a $13B Groq outlay at peak valuations raise timing-discipline questions.
what to look forHow well management deploys free cash flow across reinvestment, buybacks, dividends, and M&A. Judged by what they did, not what they said.Debt sustainability
9.5/10why 9.5A fortress balance sheet — ~$54B net cash, ~400× interest coverage, non-financial covenants — with only large off-balance-sheet supply commitments as a caveat.
what to look forWhether the balance sheet can carry its obligations through a downturn. Coverage ratios and net debt against the cash the business throws off.Insider alignment
9.0/10why 9.0Founder-CEO holds a 3.58% stake with 96% performance-linked pay and no hedging or pledging; the only blemish is minor related-party items.
what to look forWhether the people running the business have real skin in the game, with incentives aligned to long-term owners.Cyclicality awareness
3.0/10Demand is highly cyclical and volatile across data-center AI, gaming, and crypto, with current results sitting at the peak of an AI investment wave.
what to look forHow exposed earnings are to boom-bust demand cycles, and whether the model shows through-cycle resilience rather than peak-of-cycle results.Every dimension runs this same three-pass loop: score, then a skeptic argues it down, then a judge rules. Expanded here for one; the rest stay collapsed to keep this readable.How this was scored (3-pass)
Pass 1 · initial scoring
NVIDIA's business is highly cyclical and volatile, driven by rapidly shifting demand across data-center AI buildouts, gaming, and crypto mining. The risk factors explicitly acknowledge that supply/demand mismatches have 'significantly harmed our financial results' before, and that demand spikes tied to mercurial use cases create severe forecasting problems. Revenue is concentrated in a few large customers and driven by hyperscaler capex cycles. Current results are clearly at the peak of an AI investment wave, management provides no through-cycle resilience modeling, and the 2022–2023 crypto/gaming bust confirms sharp downturns.
Counter-evidence consideredThe current AI buildout is exceptionally large and may be a longer-duration secular trend than typical semiconductor cycles, potentially moderating near-term cyclical risk. NVIDIA's CUDA software ecosystem and expanding recurring revenue could buffer pure hardware cycles. However, neither factor is modeled or quantified as through-cycle resilience.
Citations
“Significant mismatches between supply and demand have varied across our market platforms, resulted in both product shortages and excess inventory, significantly harmed our financial results and could reoccur.”
Management itself flags repeated demand/supply mismatches causing material harm, a hallmark of cyclical exposure.
“The use of our GPUs for new, mercurial, or trendy applications, has impacted and can impact in the future, demand for our products, including by leading to inconsistent spikes and drops in demand.”
Demand driven by speculative/trend end-uses is boom-bust by nature, not stable recurring demand.
Pass 2 · skeptic
adjustment 0.0Pass 1's counter-evidence already addresses the core tension: the AI buildout may be a longer-duration secular trend, but neither through-cycle resilience nor demand durability is quantified. Reading the primary sources independently confirms the assessment. The risk factors extensively document demand volatility, supply/demand mismatches, inventory provisions, customer concentration (one customer = 22% of revenue, another = 14%), non-cancellable purchase commitments, and crypto-driven demand spikes, plus the company's own statement that operating results have fluctuated and may fluctuate. No new primary-source evidence emerges that Pass 1 failed to address; a score of 3 is well-calibrated.
Pass 3 · judge
final 3.0 · no change
Pass 2 recommended no adjustment and confirmed the score is well-calibrated, citing the same evidence of demand volatility, customer concentration, and crypto-driven spikes. Both passes agree the counter-evidence (secular AI trend, CUDA moat) is already incorporated without quantified through-cycle resilience. Final 3.0, no change.
Step 2Reverse-DCF cross-check
Market-implied growth
16.6%
FCF growth the current price implies
5-yr actual growth
77.5%
5-year actual growth
The implied rate looks modest against the 5-year actual — but that history is a single AI-capex supercycle. The hard part is sustaining 16.6% through the next cyclical trough, which is why valuation scores low.
Valuation multiples
| Metric | Captured | vs 10-yr median |
|---|---|---|
| P/E | 32.1 | 4.0 |
| EV / EBIT | 37.8 | 5.1 |
| EV / Sales | 22.8 | |
| FCF yield | 0.91% |
Step 3Final verdict
NVDA is a Pass despite genuinely impressive moat and financial strength, driven by two flaws in the value framework: extreme cyclicality and a valuation that prices in near-perfection. Cyclicality scored 3.0 — the lowest possible with high confidence — reflecting NVIDIA's boom-bust history across gaming, crypto, and now AI data center, with the current cycle dependent on hyperscaler capex that can reverse quickly. At a FCF yield under 1% and EV/EBIT of 37.8x versus a 10-year median of 5.1x, the reverse DCF implies 16.6% annual FCF growth for 10 years — which sounds modest against the 5-year actual of 77.5%, but that figure is almost entirely a product of the current AI capex supercycle; sustaining 16.6% through the inevitable cyclical trough demands a structural earnings floor NVIDIA's history does not support. The business quality is exceptional; the price is not.
what the verdict meansThe final call is a holistic read across the dimensions, surfaced with its reasoning shown rather than hidden. It is a structured first pass to focus human attention, not a buy/sell instruction.