ValuFact

ValuFact turns SEC filings — the dense, 200-page documents public companies are required to file — into plain-English summaries, standardized financial data, scores and screens. Automatically, within minutes of publication, for more than 8,000 companies. It is live at valufact.com.

ValuFact is an independent venture I founded and built end to end on my own time: strategy, UX, architecture, data pipelines, the AI layer and production operations. Building it alone was deliberate. It is how I learned first-hand what AI-native product development demands — data quality, evaluation, cost control and governance — so I can set credible standards for the teams I lead.

What did I do in this project?

• Identified the gap between institutional-grade financial data tools ($20K+ a year) and what individual investors can access
• Conceived, designed and built the product end to end, from concept and packaging to architecture and operations
• Engineered real-time SEC monitoring that detects new filings within minutes of publication
• Built a 6-phase document cleaning pipeline that turns raw SEC filings (HTML, XBRL, SGML) into clean, structured text — cutting file size 60–85% while preserving the meaningful content
• Designed filing-type-specific AI prompt architectures with rigid JSON output schemas for consistent analysis across 10-K, 10-Q, 8-K, Form 4, 20-F and 6-K filings
• Developed a 4-layer financial concept classification engine mapping 17,738 GAAP concepts to standardized statement line items
• Built an 18-ratio analysis engine and six composite scores with data-quality tracking, trend history and plain-English explanations
• Designed four-tier subscription packaging with usage-based AI cost controls
• Built model governance into the product: pre-registered acceptance gates, masked holdouts, immutable prediction ledgers and a configuration audit trail
• Run the production stack — FastAPI, PostgreSQL and a systemd worker fleet — with encrypted off-box backups

The Problem

Every quarter, thousands of publicly traded companies file detailed financial reports with the SEC. These filings are the single most authoritative source of truth about a company’s financial health — but almost no one can actually use them.

• They’re enormous. A typical 10-K filing runs 200+ pages of dense legal and financial text.
• They’re written in accounting jargon. The same metric might appear under 17,000+ different naming conventions across companies.
• They’re inconsistent. Apple’s balance sheet doesn’t look like Goldman Sachs’s — even though they report to the same regulator.
• There’s no standard summary. Making sense of raw filings requires financial expertise, context, and hours of time.

The tools that solve this — Bloomberg, Refinitiv, FactSet — cost $20,000–$25,000 per year. ValuFact closes this gap.

AI-Powered Filing Analysis

When a company files a document with the SEC, ValuFact’s backend detects it, downloads it, and runs it through a proprietary multi-stage processing pipeline. What comes out is a structured, human-readable analysis.

What Users See

Every filing is transformed into a set of structured, readable views:

• Headline summary — One plain-language sentence on what the filing says
• What happened — The facts and figures, point by point
• Why it matters — What the results mean for the company
• Drivers and risks — What drove the results, and the risks the filing itself names
• What comes next — What to watch from here
• Section by section — Every part of the filing summarized on its own (22 sections for Apple’s 10-K)
• The numbers in this filing — Headline figures from the filing’s own XBRL against the year before, with links back to the original on sec.gov

Before & After: Raw SEC Filing → ValuFact Summary

The raw SEC filing (left) vs. what ValuFact produces (right) — the same document, transformed into structured, actionable intelligence.

Before & After: Raw Financial Data → Standardized Statements

Raw financial data buried in SEC filings (left) vs. clean, standardized, and comparable statements ValuFact produces (right).

The Multi-Stage Processing Pipeline

Every filing goes through a purpose-built 6-stage pipeline:

1. Detection — Continuous SEC RSS monitoring every 10 minutes with trading calendar awareness
2. Download — Rate-limited retrieval from EDGAR, paced to respect SEC access limits
3. Cleaning — A 6-phase document transformation: XBRL truncation, header extraction & section marking, table extraction to CSV, HTML cleaning, section detection, and normalization
4. Token Counting — Precise token measurement via API to determine optimal processing strategy
5. AI Analysis — Filing-type-specific prompt architectures with structured JSON schemas, multi-chunk synthesis for large filings, and strict handling of missing data to eliminate hallucinations
6. Storage & Delivery — Analysis is pre-computed and stored so users get instant results with zero loading delay

Financial Data Standardization

ValuFact’s 4-layer concept classification engine maps every one of 17,738 GAAP concepts to a standardized financial statement structure:

1. Manual Overrides (100% confidence) — Hand-crafted mappings for edge cases
2. FASB Presentation Linkbase (100% confidence) — Official accounting standards board guidelines
3. FASB Schema Matches (95% confidence) — Validated elements from FASB schema definitions
4. Heuristic Rules (70% confidence) — Pattern-based classification for remaining concepts

The result: every company’s financials are presented in an identical, comparable format — Balance Sheet, Income Statement, Cash Flow, and Statement of Stockholders’ Equity.

Financial Ratios & Metrics

ValuFact automatically calculates 18 key financial ratios across five categories:

• Profitability — ROA, ROE, Net Profit Margin, Gross Profit Margin, Operating Margin
• Liquidity — Current Ratio, Quick Ratio, Cash Ratio
• Leverage — Debt-to-Equity, Debt-to-Assets, Equity Ratio, Interest Coverage
• Cash Flow — Operating Cash Flow Ratio, Free Cash Flow, Free Cash Flow Ratio
• Efficiency — Asset Turnover, Inventory Turnover, Receivables Turnover

Each ratio includes historical trend visualization, period-over-period comparisons, and plain-English explanations of what the numbers mean.

Company Profiles & Explore

Users can browse and search across 8,000+ companies, filter by sector or industry, and access dedicated company hubs showing recent filings with AI summaries, standardized financial statements, ratio dashboards, and key metrics with trend charts.

What Makes ValuFact Different

• Not a ChatGPT wrapper — Purpose-built infrastructure with a 6-phase cleaning pipeline, filing-type-specific prompts, rigid output schemas, and multi-stage chunk synthesis
• Direct from the source — Every piece of data pulled directly from the SEC’s EDGAR system — no third-party aggregator
• Transparency by design — Every summary links to the original filing; every financial statement shows a coverage score
• Built for both audiences — ELI5 summaries for regular investors, standardized multi-period financials for professionals

Model governance and controls

AI is core to ValuFact, so it is governed like any other production system. Nothing reaches users because a model sounds confident; it ships when it clears a bar that was set before the test.

• Pre-registered acceptance gates — Every new analytical feature has a pass mark written down before it is measured. The cross-company Pattern Map, for example, stays admin-only until it passes a pre-registered retrodiction test on the live pipeline.
• Masked holdouts and point-in-time testing — Backtests replay scores exactly as they stood on each date. Experiment windows are fixed at creation, the holdout is masked with a one-time, logged reveal, and random null controls run alongside every sweep.
• Immutable ledgers — Predictions are written once, with their starting point and deadline, and graded by code. Nothing is re-scored after the fact.
• A full configuration audit trail — Every change to scoring and theme settings is recorded with who, why, before and after, and any version can be restored.
• Rules first, models where they earn it — Where a deterministic rule measures better than a model, the rule ships. For insider-trade intent, a rule over the SEC’s own transaction codes labelled 97% of filings at no cost, leaving a model one small, well-defined job.

Since launch

The first release summarized filings and standardized financials. Since then ValuFact has grown into a research platform:

• Six composite scores per company — Quality, Valuation, Solvency, Growth, Efficiency and Overall — recomputed nightly
• Screener, watchlists, holdings and threshold alerts built on the 18 ratios
• Market & Economy page with index proxies, the Treasury curve, economic indicators and earnings calendars
• Insider and congressional trade intelligence, with each trade’s outcome measured against the market
• Point-in-time backtesting with an anti-overfitting experiment protocol
• Four-tier subscription packaging — free, basic, professional and enterprise — with usage-based AI cost controls
• Encrypted off-box backups and an indexed archive of every stored filing

By the Numbers

• 8,000+ companies covered, 7,000+ with standardized financial data
• 17,738 concept mappings in the classification engine
• 18 financial ratios calculated automatically with trend analysis
• 6 composite scores per company, recomputed nightly
• 6 filing types supported with type-specific AI analysis (10-K, 10-Q, 8-K, Form 4, 20-F, 6-K)
• 4 subscription tiers designed around usage-based AI costs

---

Appendix: Full 10-K Analysis Example

ValuFact’s complete AI-generated analysis of Apple’s fiscal 2025 10-K, as users see it: the headline summary, what happened, why it matters, drivers and risks, what comes next, a section-by-section breakdown and the filing’s own numbers.