Engineered high-throughput "Agentic RAG" workflows using Mastra to automate sector mapping, price discovery, and organizational data enrichment without relying on static vector databases.
Situation
Manual market research across international regions, pricing synthesis, and top-down market sizing calculations was too slow to map 50+ emerging market sectors.
Task
- •Replaced static vector RAG with a dynamic "Agentic RAG" pipeline, using Tavily and Exa APIs to retrieve real-time, primary-source market data while actively filtering out aggregator domains.
- •Integrated the Vercel AI SDK to feed search context into LLMs, enforcing strict Zod schemas to guarantee structured JSON outputs for market sizes and currency-adjusted prices.
- •Created a HITL process in CMS before final data ingestion.
Result
- •Fully automated complex sector calculations and competitive price gathering, reducing strategic research time from hours to under 60 seconds.
- •Successfully mapped 50+ sectors autonomously, establishing a near-instant digital ingestion pipeline with high-confidence source citations.
My Contribution
- •Built tool-based Agentic RAG workflows bypassing the need for static vector stores.
- •Designed high-throughput data extraction using strictly enforced Zod schemas.
Key Results
90% Reduction
Workflow Automation
Dropped strategic manual research time per org/sector from hours to under a minute.
50+ Sectors
Pipeline Throughput
Processed thousands of market data points autonomously via dynamic web-based RAG.
Skills & Technologies
mastraagentic ragvercel ai sdkzodtavilyexa apitypescript