Portfolio Manager
The Complete Financial Operating System β All 17 Skills in One Stack
Key Takeaway
17 AI finance skills work as one integrated system β from data ingestion through analysis, execution, and reporting β run by 31 agents across 8 teams. This is what a complete financial operating system looks like in production.
The Problem
Financial operations are fragmented by design. Your market data is in one system. Your accounting is in another. Portfolio tracking in a spreadsheet. Risk models in Python notebooks. LP reporting in PowerPoint. SEC monitoring in an email alert you forgot to set up. Tax planning in your accountant's head.
Each tool does one thing. None of them talk to each other. The human β you β is the integration layer. You're the one pulling data from five sources, copy-pasting into a model, interpreting the output, and formatting it for the board. You are the API.
This doesn't scale. When you have a holding company with three portfolio companies, a personal investment portfolio spanning US equities, Japanese equities, French equities, crypto, and DeFi, plus fund management and LP reporting obligations β being the human integration layer is a full-time job. And it shouldn't be.
I built something different. Not a single monolithic platform. A system of 17 specialized skills, orchestrated by agents, that work together as one coherent financial operating system.
The Solution
The PyratzLabs Financial Operating System runs on Mr.Chief. 17 skills organized into four layers. Two primary agents β Hari (investment advisor) and Warren (CFO) β orchestrate everything and report to Alfrawd, the master agent.
This isn't a demo. It runs in production. Every day.
The Process
Layer 1: Data Ingestion
Everything starts with data. Four skills handle ingestion from different source types:
yamlShow code
# data-layer.yaml
skills:
alpha_vantage:
function: "Traditional market data"
covers: "US/international equities, forex, macro indicators"
feeds: [portfolio_watcher, backtest_expert, financial_analysis]
crypto_market_data:
function: "Crypto market data + on-chain metrics"
covers: "BTC, ETH, DeFi tokens, on-chain analytics"
feeds: [crypto_regime_classifier, backtest_expert, base_trader]
sec_filing_watcher:
function: "SEC EDGAR monitoring"
covers: "S-1, 10-K, 8-K, 13-F filings"
feeds: [financial_analysis, portfolio_watcher]
cmc_api:
function: "CoinMarketCap data"
covers: "Listings, categories, trending, OHLCV"
feeds: [crypto_regime_classifier, base_trader]
The data layer runs continuously. SEC watcher checks EDGAR every 30 minutes. Market data refreshes on market hours. Crypto data is 24/7. On-chain metrics update with each block.
Layer 2: Analysis
Raw data becomes intelligence. Seven skills handle different analytical functions:
yamlShow code
# analysis-layer.yaml
skills:
backtest_expert:
function: "Strategy validation and robustness testing"
input: [alpha_vantage, crypto_market_data]
output: "Validated signals with confidence intervals"
macro_regime_detector:
function: "Economic cycle classification"
input: [alpha_vantage, FRED_data]
output: "Regime state + transition probabilities"
crypto_regime_classifier:
function: "Crypto cycle phase detection"
input: [crypto_market_data, on_chain_data]
output: "Cycle phase + exit/entry signals"
financial_analysis:
function: "Company-level fundamental analysis"
input: [alpha_vantage, sec_filing_watcher]
output: "DCF models, ratio analysis, earnings previews"
monte_carlo_simulator:
function: "Probabilistic scenario modeling"
input: [any_numerical_data]
output: "Distribution of outcomes with confidence intervals"
risk_metrics:
function: "Portfolio risk quantification"
input: [portfolio_watcher, alpha_vantage]
output: "VaR, CVaR, Sharpe, Sortino, drawdown analysis"
argus_edge:
function: "Prediction market analysis"
input: [polymarket_data, kalshi_data]
output: "Calibration reports, EV calculations, bet sizing"
Layer 3: Execution
Analysis drives action. Three skills handle execution:
yamlShow code
# execution-layer.yaml
skills:
portfolio_watcher:
function: "Real-time portfolio monitoring + alerts"
input: [all_data_sources, all_analysis_skills]
output: "Position alerts, rebalance recommendations"
agent: Hari
base_trader:
function: "Crypto trading on Base via Bankr"
input: [crypto_regime_classifier, backtest_expert]
output: "Trade execution (with approval gates)"
agent: Hari
# DeFi scanner (in development)
defi_scanner:
function: "DeFi opportunity detection"
input: [crypto_market_data, on_chain_data]
output: "Yield opportunities, arbitrage signals"
agent: Hari
Every execution action requires approval. Hari recommends. I decide. The agent never trades autonomously β that's a design principle, not a limitation.
Layer 4: Reporting & Compliance
Numbers become documents:
yamlShow code
# reporting-layer.yaml
skills:
xbrl_filing_generator:
function: "Structured financial statement tagging"
input: [financial_statements]
output: "XBRL-tagged filings, PCG-to-IFRS conversion"
agent: Warren
financial_statements:
function: "Income statement, balance sheet, cash flow"
input: [accounting_data]
output: "GAAP/IFRS-compliant statements with variance analysis"
agent: Warren
startup_financial_modeling:
function: "Runway, revenue, and growth modeling"
input: [monte_carlo, accounting_data]
output: "Board presentations, LP reports, scenario analysis"
agent: Warren
gaap_ifrs_compliance:
function: "Standards compliance validation"
input: [financial_statements, xbrl_filings]
output: "Compliance flags, remediation guidance"
agent: Warren
Agent Orchestration
The agents aren't just skill runners β they're domain experts:
View details
HARI (Investment Advisor)
βββ Morning brief: Market regime + portfolio positions + signals
βββ Monitors: All data sources continuously
βββ Recommends: Allocation changes, entries, exits
βββ Reports to: Alfrawd β Bilal
βββ Never: Executes without approval
WARREN (CFO)
βββ Quarterly: Financial statements + LP reports + board deck
βββ Monitors: Cash flow, runway, compliance deadlines
βββ Handles: PCG-to-IFRS, XBRL tagging, audit prep
βββ Reports to: Alfrawd β Bilal
βββ Never: Signs or files without approval
ALFRAWD (Master Agent)
βββ Synthesizes: Cross-domain intelligence
βββ Escalates: Conflicts between Hari and Warren's priorities
βββ Delivers: Consolidated briefs to Bilal
βββ Coordinates: 31 agents across 8 teams
A Typical Day in the System
View details
06:00 UTC β Crypto data refresh (overnight moves)
06:15 UTC β Crypto regime classifier updates cycle phase
06:30 UTC β Hari generates morning intelligence brief
07:00 UTC β SEC watcher checks for overnight filings
07:30 UTC β Macro regime detector ingests pre-market data
08:00 UTC β Hari delivers morning brief to Alfrawd β Bilal
09:30 UTC β US market open: portfolio watcher activates
10:00 UTC β Alpha Vantage data flowing, positions updating
12:00 UTC β Mid-day risk metrics calculation
16:00 UTC β US market close: daily P&L summary
16:30 UTC β Backtest expert runs daily signal check
17:00 UTC β End-of-day portfolio snapshot
21:00 UTC β Warren runs any scheduled financial reports
22:00 UTC β Overnight monitoring: crypto + SEC + news
The Results
| Metric | Before (Manual) | After (17-Skill Stack) |
|---|---|---|
| Data sources monitored | 2-3 (when I remembered) | 8+ (continuous) |
| Portfolio check frequency | Daily (manual) | Real-time |
| Earnings prep time | 15-20 hours/quarter | 10 minutes |
| Financial statement prep | 3-4 weeks | 2-3 days (including review) |
| Regime detection lead time | Reactive (news-based) | 2-3 weeks proactive |
| LP report generation | 1-2 weeks | 1 day (draft + review) |
| PCG-to-IFRS conversion | β¬18K + 3 weeks | 5 minutes + 45 min review |
| SEC filing detection | When I saw the news | < 30 minutes after filing |
| Risk metrics calculation | Quarterly | Daily |
| Prediction market analysis | Gut feel | Calibrated, data-driven |
| Human time required | 30-40 hours/week | 5-8 hours/week (review + decisions) |
The 30-40 hours to 5-8 hours reduction isn't because the work disappeared. It's because the work shifted. I went from doing the work to reviewing the work. From generating reports to reading reports. From collecting data to interpreting data.
That's the real transformation. I'm not faster at the same job. I have a different job. My job is now judgment, not execution. Decision-making, not data-gathering.
What This Actually Costs
Let's be honest about the investment:
- Setup time: ~2 weeks to configure all 17 skills
- API costs: Alpha Vantage, CoinMarketCap, Glassnode, Apify (~$200/month)
- Mr.Chief compute: Agent runtime (~$150/month)
- Maintenance: 2-3 hours/month updating configs and reviewing accuracy
- Total monthly cost: ~$400
- Total monthly time saved: ~100 hours
The ROI isn't close. And it compounds β every month, the system gets better calibrated, the backtests get deeper, the calibration data gets richer.
Try It Yourself
You don't need all 17 skills on day one. Start with the layer that matters most:
If you're an investor: Start with Alpha Vantage + SEC Filing Watcher + Portfolio Watcher. That's your data + monitoring foundation.
If you're a CFO: Start with Financial Statements + Startup Financial Modeling + Monte Carlo. That's your reporting + planning foundation.
If you're in crypto: Start with Crypto Market Data + Crypto Regime Classifier + Backtest Expert. That's your analysis foundation.
Add skills one at a time. Let each one prove its value before adding the next. The integration happens naturally through Mr.Chief β skills that produce data are automatically available to skills that consume data.
The full stack took us 6 months to build and tune. It was worth every week.
A 31-agent financial operating system sounds complex. Running it is simple. That's the whole point.
Related case studies
Portfolio Manager
Portfolio P&L Across Stocks, Crypto, and DeFi β One Message, Every Day
Unified portfolio tracking across US equities, Japanese stocks, French stocks, crypto, and DeFi positions. One daily Telegram message with P&L, allocation drift, and benchmark comparisons.
Portfolio Manager
Earnings Preview Report for 5 Holdings β Before the Market Opens
How an AI agent generates earnings preview reports for NVDA, GOOGL, META, TSLA, and PLTR in 10 minutes β consensus estimates, implied moves, and key metrics delivered before Monday open.
Portfolio Manager
Forex Impact on Our Portfolio β The EUR/USD Move Cost Us β¬47K
AI agent tracks daily FX impact on a multi-currency portfolio spanning EUR, USD, JPY, and GBP. Separates investment returns from currency effects for board reporting. Mr.Chief.
Want results like these?
Start free with your own AI team. No credit card required.