Unpacking an Agentic Workflow
summary
A working agentic workflow routes user queries to small domain agents, uses Angular front end, Python backend, LangChain elements plus custom Python, and calls an SQL-writing LLM tool for transaction queries and a tiny RAG for product knowledge; calculations move to functions to avoid LLM math errors. Real-time transaction enrichment and budget planning power answers, accuracy tested with automated scenarios (run 100×) and human evaluation, typical latency is 5–8s simple 20–30s complex, and maintenance and evaluation take longer than initial development.
AI-written summary · about
chapters (22)
- 0:00Agentic workflows in finance
- 2:14Financial assistants need agentic workflows
- 5:54Limitations of single-agent solutions
- 8:25Financial assistant UI demo
- 10:13Agents route and process queries
- 14:37UX formatting of financial insights
- 18:38Data sources for financial assistant
- 21:09Agentic workflow architecture
- 24:22Separating LLM and non-LLM tasks
- 27:34Financial literacy and accuracy challenges
- 31:30Agent memory: short-term vs long-term
- 33:21Scalability and response time
- 35:27Cost and development challenges
- 37:38Q&A: Risks and data integration
- 39:28Q&A: Privacy and model selection
- 41:12Q&A: Testing accuracy and malicious use
- 43:02Q&A: Framework and model choices
- 44:54Q&A: Handling financial irregularities
- 46:04Q&A: AI improves financial literacy
- 47:08Q&A: User experience metrics
- 47:50Q&A: Response time for complex queries
- 49:19Q&A: Managing topic switching
related talks
35mAgentic Workflows: Stories from the Engine Room
50mAI Agentic Workflows & Design Patterns
59mOptimizing Complex Workflows with Event - Driven - Multi Agentic Approach
51mOptimizing Complex Workflows with Event-Driven Multi-Agentic Approach
22mAgentic AI for Engineering Leaders | Strategy & Execution Guide
50mPractical Agentic RAG with SpringAI in Modern Enterprise Applications
01h 08mMigrate and Modernize Legacy Workflows to Intelligent Agents (PEX304)
31mBuilding Agentic AI Platforms
