All work
FinanceAI AgentsAI-native

Weekly Portfolio Analyzer

Scheduled agent that turns a brokerage export into a Gamma deck, with its own visual QA loop.

Year 2026Status BuiltType Artifact
16
step pipeline
3
parallel subagents
auto
visual QA loop
weekly
scheduled run

Overview

An automated weekly portfolio-analysis agent. It reads a brokerage portfolio export, runs deterministic quantitative analysis in Python (GICS-standard ETF look-through sector allocation, risk scoring, performance vs. benchmarks), and generates a Gamma presentation through a 16-step pipeline. Three subagents (benchmarks, macro, thesis) run concurrently; a Claude-in-Chrome visual-QA loop screenshots slides and logs format issues back into a YAML feedback file. Runs on a weekly schedule with PDF export.

Highlights

  • 16-step pipeline: classification → metrics → scoring → deck → visual QA
  • Three parallel subagents for benchmarks, macro context, and thesis
  • GICS-standard ETF look-through sector exposure
  • Automated screenshot-based visual QA with a self-maintaining issue log

Stack & tools

  • Claude Cowork
  • Python
  • YAML
  • Gamma (MCP)
  • Claude-in-Chrome
  • Pytest
  • Excel / CSV
Weekly Portfolio Analyzer · deck preview

Architecture

A scheduled agent that turns a raw brokerage export into a finished deck. It classifies holdings with GICS-standard ETF look-through, runs deterministic risk and performance math in Python, then fans out to three concurrent subagents for context. A Claude-in-Chrome loop screenshots the generated slides and writes formatting issues back into a YAML log that the next pass reads, a self-correcting QA cycle most decks never get.

Brokerage exportCSV / XLSX
Classify
GICS look-through allocation
Parallel subagents∥ concurrent
Benchmarks
Macro
Thesis
Compute
Risk scoring + performancePython
Deck + QA
Gamma deck
Visual QA loopClaude-in-Chrome → YAML
Scheduled PDF
InputAgentEngineGateOutput