RedInk.

The problem

Junior and mid-level equity research analysts at small-to-mid investment firms cover dozens of companies per quarter but can't afford Bloomberg Terminal and don't have time to manually cross-reference every MD&A claim against the financial statements. Management spin detection — noticing when language and numbers diverge — is core to good analysis and structurally under-tooled outside of frontier institutions.

The downstream pain stack: time vs coverage (you can't read 30 filings a day, so you miss the one that mattered), trust in AI tools for reliable financial analysis is low (one hallucinated number kills the workflow forever), and management language is hard to decipher (deliberately vague to soften bad signals).

The solution

RedInk gives the analyst a structured starting point and a trust-calibrated answer engine. The Anomaly Leaderboard ranks QQQ companies by anomaly score for the latest quarter so the analyst knows where to start. The Company-Quarter Detail view shows what's driving the anomaly (margin contraction + liabilities jump, for example) with the math visible.

The Ask a Question flow handles freeform analytical follow-up ("what's driving the anomaly vs last quarter?") and returns answers with citations and an Evidence Panel showing the SEC source line items the answer is grounded in — plus an Eval Panel showing the LLM-as-judge result so the analyst can calibrate trust per-answer rather than per-product.

How it works

Coverage scope: NASDAQ QQQ universe, 5 years of earnings announcement data. Flow A ("Show me something interesting"): analyst lands on the anomaly leaderboard → clicks a company-quarter card → sees anomaly score + top drivers + time-series chart → asks "what drove this anomaly?" → gets a computed answer with citations and grounded reasoning.

System prompt explicitly frames the AI: "You are RedInk, a financial anomaly analyst for QQQ-universe companies. Your job is to explain what changed in a company's financials quarter over quarter, why it matters, and whether it warrants attention. You are skeptical by default. You work like an analyst, not a chatbot." Refusal cases: questions about the future are refused (the product is anomaly detection, not forecasting).

Who it's for

Primary external users: junior and mid-level equity research analysts at small-to-mid-size investment firms who don't have Bloomberg Terminal budgets but still need to cover the QQQ universe at depth. Budget authority + urgent pain + most to lose from missing something buried in a 10-Q make them the buyer-and-end-user collapsed into one segment.

Secondary: equity research teams, hedge fund analysts, audit firms, and compliance departments as the B2B revenue layer at scale. Free tier: retail investors and finance enthusiasts looking for quick anomaly insights — useful for top-of-funnel growth even though they won't convert.

Why it matters

AI-powered financial tooling is in a 30%+ growth wave, and incumbents like Bloomberg Terminal are structurally exposed: they're expensive, dominant, and sticky — but they're also generalist tools competing in an era of specialized AI-native analytics products that beat them on specific high-value workflows.

RedInk's structural bet is focused vertical depth + visible trust mechanics. The product doesn't try to replace Bloomberg; it does one thing very well (management spin detection in QQQ filings) with the trust scaffolding (citations, evals, computations) that makes analysts willing to use AI for their day job. Once trust is established on the narrow workflow, expansion to the broader research stack becomes credible.

At a glance

Project

RedInk

Built by

Sheldon Gomes

One-liner

AI financial anomaly detector for NASDAQ-100 — spots when management narrative diverges from the financials, with computed drivers, SEC citations, and visible trust scores.