The Challenge
An investment group was employing analysts to simply monitor and summarize over 50 disparate newsletters and corporate blogs daily. It was tedious, prone to human error, and expensive enough that coverage gaps regularly occurred.
Implementation
Tool-calling agents reading and reasoning in the background.
LangGraph Stateful ExecutionBuilt a complex Directed Acyclic Graph (DAG) state machine to govern the scraping, parsing, summarization, and embedding phases linearly.
Headless Scraping Tool CallingEmpowered the OpenAI agent to explicitly use Exa and Playwright tools to dynamically bypass anti-bot challenges and fetch the content.
Supabase & pgvector RAGPushed the structured synthesized data into Supabase, where pgvector enables instant semantic similarity queries across years of scraped knowledge.
Ongoing Agent OpsSupported via the Agent Ops retainer to continually adjust DOM selectors and prompt instructions as scraping targets mutate.
50+
Data sources mapped implicitly
~2K
Articles synthesized monthly
1.4s
pgvector semantic retrieval
Repetitive tasks taking up your time?
We build custom AI agents that execute complex backend workflows perfectly, 24/7 without fail.