Anant Shankhdhar is a machine learning engineer at RISA Labs developing evidence-grounded oncology automation for insurance decisions that can delay cancer treatment. His systems combine patient records, payer rules, language models, and clinician oversight to determine when prior authorization can proceed automatically.
Shankhdhar studied computer science and engineering at the Indian Institute of Technology Guwahati. In 2022, he coauthored XInfoTabS, a ten-language benchmark testing whether information in a table supports, contradicts, or leaves a claim unresolved. He maintains the project website, and its open-source dataset and translation code support further research into multilingual tabular reasoning.
His earlier experience includes the University of Utah, Pibit.ai, and Adobe Research’s Multimodal Content Lab. At Walmart Global Tech, he worked as a data scientist on retrieval-augmented generation, multilingual systems, and conversational AI before applying related techniques to oncology workflows at RISA Labs.
Automating consequential decisions safely
- Four-agent prior-authorization architecture: Separate agents verify insurance eligibility, classify authorization requirements, assess medical necessity, and submit supporting documentation.
- Deterministic checks and self-healing automation: A coverage orchestrator normalizes payer API and portal responses, applies explicit eligibility rules, and uses model-generated configurations to repair fragile portal automations.
- Multi-source authorization corroboration: Historical authorization letters, portal checks, and a SQL-backed payer-rule knowledge base establish whether approval already exists or is unnecessary, allowing corroborated cases to advance without manual review.
- Confidence-based clinical escalation: A patient biomarker knowledge graph, clinical notes, and insurer criteria inform medical-necessity decisions; uncertain or contradictory cases return to clinicians.
His oncology workflow architecture makes automation conditional on traceable evidence and preserves professional judgment where the available information cannot support a confident decision.
