Quant Research Platform
Six data sources. 10TB of financial data. Same underlyers in different schemas. Vyuh extracted the ontology from each, reconciled the entities, and built a unified semantic model that AI agents operate on — without knowing anything about the underlying plumbing.
See HowSame concepts, six different schemas
Quantitative research requires data from many sources — each with its own schema, identifiers, granularity, and access patterns. The same underlyer (AAPL, TSLA, SPY) appears in every source but is represented differently in each. Traditional approaches duct- tape these together with brittle ETL pipelines and manual mapping.
Stock Data
Tick-level OHLCV, sub-second granularity, vendor-specific formats
Options
Chains, Greeks, implied volatility — different structure per exchange
Fundamentals
Financial statements, ratios, estimates — varying fiscal calendars
Earnings
Calendars, transcripts, surprises — event-driven, unstructured
News & Sentiment
Real-time feeds, NLP scores, entity tags — no shared schema
Short Interest
Bi-monthly snapshots, settlement-date keyed, broker-specific
Cross-source entity resolution
Connecting to data sources is the easy part. The hard part: recognizing that the same business concepts appear in different forms across sources — and reconciling them into a single truth.
The Problem
AAPL in stock data uses CUSIP. Options use OCC symbology. Fundamentals use CIK. News uses entity names.
The Resolution
Vyuh resolved all representations into a single underlyer entity with canonical identifiers across all six sources.
The Problem
Tick data is sub-second. Fundamentals are quarterly. Earnings are event-driven. News is real-time.
The Resolution
Vyuh built a temporal alignment layer — agents query by business time, and the semantic model routes to the right granularity per source.
The Problem
Stock price is a trade. Options price is a quote. Fundamental "price" is a ratio denominator. Each has different meaning.
The Resolution
Vyuh disambiguated price semantics by context — agents request typed price capabilities, not raw numbers.
30+ governed capabilities from a unified model
The reconciled ontology became a governed capability layer. Each capability is typed, validated, and permissioned — AI agents discover and use them without knowing anything about the underlying sources.
Market Data
- OHLCV prices
- Tick data
- Splits & dividends
- Market hours
Options
- Options chains
- Greeks
- Implied volatility
- Open interest
Fundamentals
- Financial statements
- Ratios
- Analyst estimates
- Insider trading
Analytics
- Technical indicators
- Momentum scores
- Volatility models
- Correlations
Events
- Earnings calendar
- Earnings transcripts
- News feed
- SEC filings
What we learned
Entity resolution is the real work
Connecting to data sources is easy. Understanding that the same underlyer appears in six different schemas with six different identifiers — and reconciling them — is where the value is created.
Semantic models compound
Every new source we connected made the existing capabilities richer. Options data improved stock analysis. Earnings data improved fundamental screening. The unified model is more than the sum of its parts.
Agents don't need to know the plumbing
The AI agent doesn't know about parquet files, vendor APIs, or settlement dates. It sees typed capabilities — get_price, screen_options, analyze_earnings — and the semantic model handles the routing.
Zero rewrites as sources grow
When we added the sixth data source, the agent automatically gained new capabilities. No code changes. No retraining. The governed model grew, and the agent grew with it.