A collection of production-ready AI projects — from multi-agent office automation to AI-powered ticket management and investment analysis frameworks.
Each project tackles a real workflow pain-point using LLMs and autonomous agents — designed to be deployed, not just demoed.
AI-powered task ticket manager that analyses a software project (local repo or GitHub URL) and automatically generates structured, actionable tickets — no database required. Think Jira, but driven by an LLM that reads your code.
Multi-agent office automation built on CrewAI and GPT-4o. Handles email drafting, document summarisation, and US stock analysis — all scheduled automatically and tracked in a Streamlit dashboard with SQLite cost logging.
Professional investment analysis frameworks for US stocks — works on every AI platform. 21 structured prompt skills turn any LLM into an institutional-quality analyst. No runtime; just prompts that plug into Claude Code, Cursor, Gemini CLI, or ChatGPT.
Four specialised CrewAI crews run independently via APScheduler. Every run is persisted to SQLite so you can audit token usage and cost.
Drafts email body from natural-language intent, then sends via Gmail SMTP using GmailSendTool.
Reads PDF / DOCX / TXT → produces a structured summary → emails it automatically.
Pulls live yfinance data, runs GPT-4o analysis, delivers a market summary to your inbox.
Every crew run logs token count, cost, status, and timing — visible in the Streamlit dashboard.
main.py (CLI)
├── email → EmailCrew → Gmail Agent → GmailSendTool
├── summarize → DocSummaryCrew → Document Analyst → DocReaderTool → GmailSendTool
├── stock → StockCrew → US Stock Analyst → yfinance → GmailSendTool
├── scheduler → APScheduler (reads schedule.yaml)
└── ui → Streamlit Dashboard
All agents → LLM: OpenAI GPT-4o → runs.db (SQLite)
Every framework is a battle-tested prompt that works as a slash command in Claude Code or as a plain-text input on any LLM. No installation for most platforms.
Piotroski score, ROIC, moat rating, FCF yield, buy / hold / sell signal in one shot.
Identify rotation opportunities and relative strength across S&P 500 sectors.
Track SEC Form 4 filings — spot clusters of insider buying before they move.
Sentiment, key themes, and management tone extracted from earnings transcripts.
Track 13F filing changes — follow smart-money moves across hedge funds.
Evaluate allocation, risk-adjusted returns, and concentration for any portfolio.
# Claude Code — recommended, full slash-command support
claude
/plugin marketplace add yennanliu/InvestSkill
/plugin install us-stock-analysis
/us-stock-analysis:stock-eval AAPL
# Cursor / Gemini CLI / ChatGPT — zero install
cat prompts/stock-eval.md | pbcopy
# → paste into any LLM
Point it at a repo, get structured tickets. The LLM reads your code, infers missing context, and writes tickets an engineer would actually act on.
Feed a local path or a GitHub URL — the agent clones, indexes, and reads the codebase automatically.
LLM produces structured tickets with title, description, acceptance criteria, and priority.
Clean REST API served by uvicorn — integrate with Jira, Trello, or any issue tracker via webhook.
pytest suite covering API endpoints and storage layer — designed to run in CI from day one.
# Quick Start
uv sync --extra dev
cp .env.example .env # add OPENAI_API_KEY
uv run uvicorn main:app --reload
open http://localhost:8000
# Tests
uv run pytest tests/ -v
I'm always open to collaborating on ambitious agent projects.
Reach out or explore more of my work below.