Streamlining Investment Decisions with Python, LangChain, and RamaLama
summary
Prototype multi-agent trading pipeline built with Python, LangChain and RamaLama on Fedora aims to emit three daily stock picks but currently fails to fully automate execution. Real-time data is the bottleneck: yfinance and free Polygon limits force reliance on delayed quotes and Reuters RSS for sentiment. Architecture uses containerized agents via Podman, an S3-backed RAG knowledge base, Granite/Mistral models, and plans for Kubernetes; memory, OCI model pulls and networking are the main unresolved issues.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:47Investment committee workflow
- 3:56Prototyping agents
- 6:26Agent communication
- 7:33Data access challenges
- 9:01Open-source models
- 10:34Testing strategies
- 12:04Data source integration
- 13:59Current trading approach
related talks
16mEffective agent design patterns in production
47mRamaLama: Making working with AI Models Boring
37mWhat's new with RamaLama
41mBuild Next-Generation Applications with Language Models using LangChain and Python
23mBuilding Production Multi-Agent RAG Systems on Serverless AWS
48mUnlocking LLM Potential: From Exploration to Integration
41mThree agents, three frameworks, one talk
35mAdvanced RAG Techniques with LlamaIndex