Summary
At the AI Engineer conference, Anant Shankhdhar from Risa Labs (operating as Trisca) presents how the company built a four-agent pipeline to automate oncology prior authorization workflows — a process where human review has traditionally been required before cancer drug orders are submitted to insurers. The goal was to identify which orders could bypass human review entirely and flow straight to submission with high confidence.
The system is built around four specialized agents: an EV agent for eligibility and benefits verification, an auth agent that determines whether a specific drug requires authorization, a necessity agent that functions as the clinical reasoning layer for cases that do require auth, and a submission agent. A central challenge is that insurance eligibility data is scattered across dozens of portals, APIs, and documents. Risa addressed this by building a unified coverage orchestrator that normalizes payer data into a consistent format and feeds it into a deterministic decision engine that flags clear-cut cases early, before LLM involvement.
To remove humans from the loop for a meaningful share of orders, the team developed a payer rule knowledge base — built from automated portal checks and LLM extractions — that reconciles multiple evidence sources before asserting authorization status with confidence. When two independent sources (authorization letters plus payer rules) agree, the system proceeds without review. Shankhdhar also covers a self-healing RPA loop that detects and mitigates broken portal automations in production without manual intervention.
📺 Source: AI Engineer · Published July 20, 2026
🏷️ Format: Workflow Case Study
