~/work / stockinsight
SauravSK1011/StockInsightopen source
StockInsight
An LLM stock advisor grounded on fresh news. It scrapes the latest headlines for a ticker and asks a local Llama 3.2 for a short buy/sell view.
- llm-apps
- news-grounding
- local-llm
01 / problem
Why it needed to exist
Retail investors drown in headlines. Can an LLM turn today’s news about one stock into a short, opinionated read?
02 / what i built
The approach
01
Fresh context
Scrapes current Moneycontrol news for the searched stock, so the model reasons over today’s headlines and not its training data.
02
Local-first LLM
LangChain prompt chain running Llama 3.2 locally through Ollama, with an OpenAI path and LangSmith tracing for debugging.
03 / architecture
How data flows
- 00Tickeruser query
- 01ScrapeMoneycontrol headlines
- 02PromptLangChain template
- 03LLMLlama 3.2 via Ollama
- 04View50-word buy / sell call
04 / results
What it did in the real world
- Open source on GitHub
05 / limitations
Honest notes
- The output is forced to buy or sell, with no hold option and no backtesting. It’s a learning project in grounding LLMs on live data, not financial advice.