← All projects
Case studyBridgeweave

Stock analysis assistant

Stock analysis and investor education through an LLM agent.

A calculator and financial documents beside a laptop
Photo by Jakub Żerdzicki / Unsplash ↗

At Bridgeweave, we were also building an assistant to help people understand stocks. Someone could ask a question and get help with the analysis, or learn about what they were looking at.

The difficult part was keeping the answer grounded. With financial information, a convincing answer still needs something behind it.

Give the agent something to work with#

I worked on fine-tuning, retrieval & tool use. We analyzed the query, used Qdrant to retrieve relevant information, and gave the agent tools to work with.

That meant the response could draw on information available to the system. Guardrails were another part of keeping it on track.

01 / Answering a questionBring the information into the answer.
  1. Investor's questionAnalyze what they're asking
  2. Retrieval & toolsRelevant information from Qdrant & tool use
  3. Grounded responseAnalysis or explanation, with guardrails

Analysis & education#

The assistant had both jobs. It helped people analyze a stock and understand the information involved in that analysis.

Fine-tuning, retrieval, tools & guardrails all contributed to that. Getting them to work together was the work behind the chat window.