| name | virattt-ai-hedge-fund |
| description | Run an ai-hedge-fund-inspired multi-lens public-equity research committee using QVeris-routed data. Use for value, quality, growth, technical, sentiment, insider, and risk analysis that produces an evidence-backed research memo without brokerage access or order execution. |
AI Hedge Fund-Inspired Research with QVeris
Use the upstream ai-hedge-fund project as a methodology reference. Replace its direct financial-data dependency with QVeris and keep all analytical reasoning local.
Data boundary
Acquire every external price, fundamental, news, insider, company, and analyst input through QVeris. Require QVERIS_API_KEY for real calls. Use Discover → Inspect → Call and the capability plan in qveris.manifest.json. Do not call financial-data providers directly.
This skill is research-only. Never access brokerage accounts or place, modify, or cancel orders.
Workflow
- Resolve the issuer, listed security, exchange, currency, requested date, and evidence cutoff.
- Discover and call historical prices, financial ratios, statements, tagged news, insider transactions, and company profile capabilities.
- Record
search_id, tool_id, provider, retrieval time, fiscal period, filing date, and market timestamp.
- Build independent evidence packets for these lenses:
- quality and fundamentals;
- intrinsic and relative value;
- growth and expectations;
- technical trend and market reaction;
- news sentiment and catalysts;
- insider activity;
- portfolio and downside risk.
- Require every lens to state its evidence, counterevidence, missing inputs, confidence, and as-of date before synthesis.
- Let the risk lens set research exposure constraints and veto unsupported conclusions. Do not translate this into an executable order.
- Synthesize a committee memo that preserves disagreements instead of averaging them away.
Quality gates
- Keep reported facts, provider labels, local calculations, and model judgments separate.
- Reject mismatched currencies, periods, or duplicate securities.
- Avoid double-counting the same source across lenses.
- Do not fabricate data when a capability is missing or a call fails.
Output contract
Return scope, as-of time, QVeris trace, lens-by-lens evidence, valuation assumptions, catalyst and risk map, disagreements, confidence, missing data, and a research-only conclusion.
Validation
node scripts/qveris_candidate_runner.mjs validate .
node --test tests/smoke.test.mjs
node scripts/qveris_candidate_runner.mjs preflight .
QVERIS_API_KEY=... node scripts/qveris_candidate_runner.mjs live .
