Problem Statement Financial analysts often face inconsistent or incomplete data in bond, stock, and derivative reports. Quick decision-making under time pressure can lead to mis-hedged positions, copy-forward errors, and hidden risks. Learners struggle to understand valuation logic without hours of research. Example: A bond may list a yield inconsistent with its rating, or a stock’s P/E ratio may diverge sharply from peers. Detecting these issues manually is slow and error-prone, while traditional AI tools hallucinate calculations and provide opaque results. Over or Under solves this by providing a deterministic, multi-model financial valuation engine that compares peer benchmarks, flags inconsistencies, and explains reasoning — all without hallucinations, in seconds.
Inspiration The recent GLD (Gold ETF) crash highlighted how market stress exposes fragile assumptions in financial documents. We wanted a tool that: • Automatically flags inconsistencies and outliers • Provides auditable, deterministic calculations • Teaches financial reasoning and valuation logic This led to Over or Under, designed for analysts who need fast, accurate pre-screening and learners who want hands-on experience with valuation models.
What it does Over or Under is a professional-grade financial valuation platform for stocks, bonds, and derivatives: • Deterministic multi-model valuation: DCF, Graham Number, Gordon Growth, CAPE, Buffett Indicator, Tobin’s Q, Fed Model, Rule of 20, PEG Ratio, Black-Scholes • Peer benchmarking: Compares P/E, P/B, EV/EBITDA, ROE, revenue growth, and more • Scoring: Outputs Overvalued, Fairly Valued, or Undervalued classifications • Explanations: AI (Claude API) generates human-readable reasoning only • For analysts: Detect inconsistencies, verify assumptions, and save hours of manual research • For learners: Explore multi-factor valuation, peer comparisons, and market metrics interactively Key differentiator: Unlike AI-only financial tools, Over or Under is deterministic, transparent, and auditable, providing both speed and educational value.
How we built it • Specialist agents: • Bond007 → fixed income • Stonker → equities • CallMeMaybe → options • Pipeline: Data → Intrinsic & Market Calculations → Peer Comparison → Multi-Factor Scoring → Verdict → Explanation • Tech stack: Python 3.9+, Streamlit, Pandas, Plotly, Claude API for explanations • Performance: Core calculations <100ms; full analysis including explanations <2s • Data: Demo dataset of 126 companies, 17 bonds, 10 options, benchmarked against sector averages The architecture separates deterministic computation from AI reasoning, eliminating hallucinations while remaining interactive, fast, and interpretable.
Challenges we ran into • Aggregating multiple valuation models into a coherent weighted score • Implementing bond and derivative analytics correctly (yield spreads, term structures, Greeks) • Maintaining speed and interactivity while ensuring deterministic correctness • Designing a system usable for both analysts and learners without oversimplifying metrics
Accomplishments that we’re proud of • 100% accuracy on 39 test assets (12 stocks, 17 bonds, 10 options) against expert consensus • Successfully scored 126 companies across sectors, outperforming AI agents prone to hallucinations • Built a multi-agent, multi-model valuation engine from scratch • Speed: Core calculations <100ms, full analysis <2s, enabling batch evaluations • Transparency: All outputs are auditable and traceable • Fully interactive hackathon demo via Streamlit • Designed for both analysts and learners, combining actionable insights with educational value
What we learned • Deterministic computation is essential in high-stakes finance — AI alone is insufficient • Separating calculation from explanation builds trust and prevents hallucinations • Multi-model valuation and peer benchmarking highlight inconsistencies and reduce bias • Learners benefit from hands-on exploration of real financial metrics, while analysts gain speed and verification
What’s next for Over or Under • Integrate real-time market data (Alpha Vantage, Polygon.io) • Expand universe to 500+ equities, 200+ bonds, more options • Multi-stage DCF with terminal value scenarios • Portfolio-level analysis and backtesting engine • Regime-aware valuation adjustments (bull/bear cycles) • Enterprise API for programmatic access and regulatory-ready reporting
Log in or sign up for Devpost to join the conversation.