Agentic Workflow
Agentic Portfolio Analyst
An automated portfolio intelligence engine designed to eliminate emotional bias from capital allocation decisions.
Utilizing GPT-5's multimodal vision capabilities, this workflow parses unstructured data (portfolio screenshots) via an OCR pipeline, mapping current holdings against a proprietary 'Q-Score' algorithm.
The system autonomously calculates drift, identifies rebalancing opportunities, and generates high-frequency buy/sell directives based on strict quantitative logic rather than sentiment.
This agentic workflow, represented by the command "Workflow(StockAdvisor:$AMOUNT)" is a prototype that ingests your latest holdings from screenshots or its DB, builds a baseline, and tags each position into buckets (e.g, Tech, Innovation, Index, Energy, Real Assets, Crypto).
It expands a curated stock/etf candidate universe, applies hard eligibility screens (if desired), and scores picks with QScore (Quality 40%, Valuation 30%, Momentum 20%, Risk 10%).
Using targets determined by the user (e.g., Tech 58%, Innovation 8%, Index 18%, Energy 5%, Real Assets 6%, Crypto 5%), it prioritizes under-target buckets and allocates exactly the amount written in the command (e.g., $100) across top names (min ticket size; no adds to over-target buckets unless signals justify).
It outputs an orders CSV, a deck with slides detailing a Sector Bar, Allocation by Bucket + ticker legend, 30-yr projection dot graph, This Week’s Orders (with QScore explainer), and a Why This Pick slide (3-4 quant+qual lines per pick), plus a summary PDF.
Canonical runs log to weeklystockadvisor.db and ask if you executed the orders (updating holdings if yes).
A0 · Screenshot Ingest Agent (optional but prioritized)
If you attach holdings screenshots with the command, A0 extracts positions → builds positions_latest.csv (Ticker, Shares, Price, Value).
OCR-first (best effort), then normalization; creates a small QA summary.
If no screenshots, the flow falls back to the latest DB snapshot (canonical runs) or prior CSV baseline (tests).
A1 · Baseline Builder
Loads positions_latest.csv (or DB fallback), calculates Pre-Execution Baseline (excludes SPAXX), and tags each holding into buckets.
A2 · Universe & Eligibility Agent
Starts from your current holdings + a broadened, curated universe (Tech, Innovation/Biotech, Energy/Nuclear, Precious Metals/Real Assets, Nasdaq/Index, Crypto proxy).
Applies hard filters (e.g., price ≥ $5, basic liquidity, rough quality/sanity checks).
Honors permanent exclusions from picks (e.g., VOO/IJH/IJR/IXUS; MMFs).
A3 · Signals Agent (framework in place; live hooks ready)
Gathers or placeholders Momentum and Earnings timing context for candidates (used in rationale and tie-breaks).
Today the allocator doesn’t depend on live momentum to function, to avoid potential API costs, but the slide/rationale will reflect signal availability and is designed to plug into live data later.
A4 · QScore & Ranking Agent
Computes a composite QScore (0–100) per candidate:
Quality (40%) — ROIC, margin trend (durability and execution).
Valuation (30%) — EV/Sales/reasonableness vs peers or null (for index/trusts).
Momentum (20%) — basic up/down bias or neutral when unavailable (expandable to live).
Risk (10%) — leverage/ND/EBITDA sanity, proxy risk where applicable.
Produces a rank-ordered list per bucket.
A5 · Targeting & Allocation Engine
Uses targets based on portfolio e.g., Tech 58%, Innovation 8%, Index 18%, Energy 5%, Real Assets 6%, Crypto 5%.
Computes gaps (Target − Actual) from your baseline and prioritizes under-target buckets.
Allocates exact $AMOUNT with:
Minimum size (prevents $2 dust buys).
No adds to over-target buckets unless strong signals justify (guardrail).
Caps on Crypto/Real Assets adds unless compelling signals appear.
Picks top QScore name within each under-target bucket and sizes orders to hit the exact dollar total (last line auto-absorbs any rounding).
A6 · Orders Builder & Rationale Agent
Outputs a clean orders table (Ticker, Bucket, Price, Shares, Cost, QScore).
Writes a “Why This Pick” rationale (3–4 lines each, quant + qual):
Quant: quality/valuation snapshots (e.g., ROIC, margins, EV/Sales, QScore).
Qual: role in the portfolio (gap-filling vs. growth tilt), macro considerations, risk notes.
Adds an Orders slide QScore explainer line: “Quality 40%, Valuation 30%, Momentum 20%, Risk 10%.”
A7 · Visualization & Deck Agent
Charts (PNG):
Sector/Theme Exposure: sorted horizontal bar, groups <3% into “Other.”
Allocation by Bucket: bars (Actual vs Target) + right-side ticker legend and a standard legend box.
30-Year Projection: dot plots every 3 years: red (Bull 14%), blue (Neutral 8%), black (Bear 2%).
Deck (PPTX):
Titles set to 38pt; slide-safe layouts to avoid overflow.
Includes Cover, Allocation by Bucket (with legend), Sector Bar, 30-Year Projection, This Week’s Orders (with QScore line), and Why This Pick slides.
Summary (PDF) and Orders CSV produced alongside parsed A0 CSV.
A8 · Persistence & Post-Run Update Agent (canonical runs only)
Logs run metadata to weeklystockadvisor.db (date/time, baseline, orders, rationales, charts/deck paths).
Asks if you executed the orders. If you say yes, it updates holdings in the DB so the next baseline reflects the new shares/cost basis.
TEST runs (runs initiated by prefixing test to the command to form testWorkflow(StockAdvisor$AMOUNT) do not write to DB or increment run counters; artifacts are labeled “TEST”.