Streamlining Stock Research with Generative AI
Compressing 3 hours a day on research to minutes.
What we were actually solving
Stock analysts tasked with explaining anomalies in stock performance had to manually set thresholds, sift through news articles one by one, and summarize findings by hand, a process that was both time-consuming and prone to error. Investment bankers downstream were dependent on this slow, manual pipeline to make decisions.
How we got there
- Step 01User Research
Conducted user research with stock analysts and investment bankers to map the pain points across their workflow.
- Step 02Journey Mapping
Identified four bottlenecks: anomaly detection, news article sourcing, summarization, and follow-up Q&A.
- Step 03Proof of Concept
Built the PoC in three stages using IBM watsonx tools: anomaly detection, article extraction , GenAI summarization, and RAG-based Q&A using IBM's Granite model.
- Step 04Optimization
Solved context window limitations through LangChain-based chunking after testing spaCy and manual approaches.
- Step 05Documentation
Produced a watsonx Playbook for IBM SkillsBuild.
What shipped
An end-to-end stock anomaly analysis workflow replacing the manual process: AutoAI detects anomalies from historical stock data, Jupyter Notebook automatically sources and extracts relevant news articles, Prompt Lab summarizes events causing anomalies, and RAG-powered Q&A enables analysts to drill deeper into any finding.