Company research, automated. Hand it a stock ticker — it pulls financial data and recent news, runs them through an LLM across five analytical dimensions, and hands back a structured Markdown report with an executive summary. Runs free on Groq, Gemini, or Cerebras, or point it at any OpenAI-compatible key you already have.
A two-hour manual research task, done in a couple of minutes. This is not financial advice — see the disclaimer.
120 min → 35 min per company · $0.005–$0.02 a report · run across 70+ companies so far.
See a full example report: BlackRock (generated free on Gemini, $0.00) · Apple.
Pick a provider, paste your own API key, type a ticker. No install, nothing stored server-side — your key is used for that one request and never persisted.
Or run the same UI locally:
git clone https://github.com/changsunglim/ai-company-analysis-framework.git
cd ai-company-analysis-framework
pip install -r requirements.txt
streamlit run streamlit_app.pyA single report covering five independent modules, plus a synthesized executive summary:
| Module | Answers |
|---|---|
| Financial health | Profitability, balance-sheet strength, valuation, cash flow |
| News sentiment | Tone and themes across recent coverage |
| Competitive position | Where the company sits against its peers |
| Risk assessment | Financial, market, and headline risks |
| Growth outlook | Forward drivers and their credibility |
Output is downloadable Markdown (and JSON), with source attribution and cost/usage metadata baked into every report.
Bring your own key. The web app ships six one-click options — and a Custom card that accepts any OpenAI-compatible endpoint, so you're never locked in.
| Provider | Cost | Best for |
|---|---|---|
| Groq | Free | Fastest free tier — the recommended default |
| Cerebras | Free | Extremely fast inference |
| Gemini | Free | Roomy free tier, solid quality |
| OpenAI | ~$0.01 / report | Most reliable, no rate cap (funded account) |
| OpenRouter | Free & paid | One key, hundreds of models |
| Custom | Any | Any OpenAI-compatible endpoint — DeepSeek, Mistral, self-hosted… |
Free tiers are rate-limited; the app paces requests automatically and shows $0.00 on the report when a free provider is used.
Four stages, each independently configurable. Collectors run concurrently; preprocessing strips the payload down to only what each module needs; analysis fans out across modules with rate-limited, retrying API calls; the reporter assembles everything into Markdown.
%%{init: {'theme':'base','themeVariables':{'primaryColor':'#ffffff','primaryTextColor':'#211F1C','primaryBorderColor':'#A9790A','lineColor':'#A9790A','secondaryColor':'#F2EFE7','tertiaryColor':'#FAF8F4','fontFamily':'Helvetica'}}}%%
flowchart LR
subgraph SRC [Sources]
F["yfinance<br/><sub>financials</sub>"]
N["Google News<br/><sub>RSS</sub>"]
I["Competitors<br/><sub>peers</sub>"]
end
SRC --> C["Collect<br/><sub>async gather</sub>"]
C --> P["Preprocess<br/><sub>dedupe · chunk · tokenize</sub>"]
P --> A["Analyze<br/><sub>5 LLM modules</sub>"]
LLM(["OpenAI-compatible LLM<br/><sub>Groq · Gemini · Cerebras · …</sub>"]) -.-> A
A --> R["Report<br/><sub>Markdown · JSON</sub>"]
- Collect — Three collectors run concurrently (
asyncio.gather): financial metrics and competitor data from yfinance, news from Google News RSS. All implement a sharedBaseCollectorinterface. - Preprocess — Filter by reliability → deduplicate with normalized fingerprinting → clean → chunk with
tiktoken. The goal is to minimize tokens sent to the API. - Analyze — The LLM engine routes chunks to modules by a relevance map (financial data goes to financial analysis, not sentiment), running through an
AsyncTaskQueuewith token-bucket rate limiting, concurrency control, and exponential backoff. The executive summary is generated last, from the other results. - Report — Jinja2 templates assemble everything into Markdown with metadata and source attribution.
The CLI reads provider settings from .env (copy .env.example). Any OpenAI-compatible provider works — free examples for Groq and Gemini are in that file.
# Basic
python -m src.main AAPL
# With a company name (improves news search)
python -m src.main MSFT --company "Microsoft"
# Non-US tickers
python -m src.main 005930.KS --company "Samsung Electronics"
# Only specific modules
python -m src.main TSLA --modules financial_analysis news_sentiment
# Custom output directory
python -m src.main NVDA --output-dir ./reportsConfigure a free provider (.env)
# Groq — free, fast
LLM_API_KEY=gsk_...
LLM_BASE_URL=https://api.groq.com/openai/v1
LLM_MODEL=llama-3.3-70b-versatile
# Gemini — free
LLM_API_KEY=AIza...
LLM_BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai/
LLM_MODEL=gemini-2.0-flash
# OpenAI — paid
OPENAI_API_KEY=sk-...Hitting 429s on a free tier? Dial concurrency down with LLM_MAX_CONCURRENT=1 and LLM_MAX_RPM=10.
Everything is tunable in config/config.yaml:
analyzer:
rate_limit:
max_requests_per_minute: 20
retry_attempts: 3
exponential_backoff: true
collector:
financial:
enabled: true
news:
enabled: true
max_articles: 15ai-company-analysis-framework/
├── streamlit_app.py # Web UI (the live app)
├── src/
│ ├── main.py # CLI entry point
│ ├── pipeline.py # Pipeline orchestrator
│ ├── collector/ # Financial · News · Industry collectors
│ ├── preprocessor/ # Dedup, clean, chunk
│ ├── analyzer/ # Async LLM engine + prompt templates
│ ├── reporter/ # Markdown / JSON report generator
│ └── utils/ # Rate-limited task queue, logging
├── config/config.yaml
├── examples/ # Full sample reports
└── tests/
- Fork or push this repo to your own GitHub.
- Go to share.streamlit.io → New app.
- Point it at your repo, branch
main, main filestreamlit_app.py. Deploy.
No secrets to configure — every visitor enters their own key at runtime, and nothing is stored. Streamlit Community Cloud redeploys automatically on each push to main.
python -m pytest tests/ -vThis tool generates analysis using AI language models and public data sources, for informational purposes only. It does not constitute financial, investment, legal, or tax advice, and no responsibility or liability is accepted for any financial decisions or outcomes based on its output. Verify data independently and consult a licensed financial advisor before making investment decisions.
Isaac Lim — CS at the University of Liverpool. GitHub
If this saved you time, a ⭐ on the repo is appreciated.