Sai Daya Shankar
Giridhar

• Built automated and human-in-the-loop evaluation pipelines, benchmarks, and quality metrics (accuracy, relevance, safety) for multi-agent LLM systems, validating agent responses against expected outputs before each release.
• Reduced hallucination rates by 70% by establishing LLMOps evaluation practices: prompt/version management, automated scoring, regression test sets, and root-cause analysis of recurring failure modes.
• Integrated continuous evaluation and monitoring into CI/CD, with distributed tracing and observability, to catch quality regressions across staging and production before they reached users.
• Designed real-time, multi-agent conversational AI (clinical QA agents, AI Health Scribe, and differential-diagnosis CDSS) over live doctor–patient dialogue and FHIR/EHR data, cutting clinician documentation time by 40%.
• Implemented HIPAA-aligned guardrails, PII/PHI handling, and bias detection across clinical agents to keep model behavior safe, fair, and compliant.
• Led and mentored an AI/ML team, set evaluation standards and best practices, and partnered with product and engineering to take agents from prototype to scale.

Hikigai Platform

• Led architecture and development of a HIPAA-aligned platform for deploying healthcare AI agents, MCP servers, and full-stack AI applications, including an Agent Registry, MCP Gateway, and Workflow Orchestration Engine for agent registration, versioning, deployment, and multi-agent composition.
• Developed SDK-based developer workflows in Python and Node.js, a FastAPI backend, a Next.js portal, and YAML/JSON deployments with tenant-aware access control and FHIR/EHR interoperability.
• Embedded continuous evaluation and runtime monitoring into the platform, running automated HIPAA and PHI compliance checks on agent outputs in real time, with distributed tracing and observability across staging and production.
• Integrated these compliance checks as evaluation gates in the platform’s CI/CD flow, blocking PHI leaks and non-compliant agent responses before they reached production.

• Engineered an infrastructure orchestration platform using cloud Infrastructure-as-Code to automate provisioning, reduce manual operations, and lower infrastructure cost.
• Built operational dashboards that surfaced company-wide performance metrics and improved visibility into business and infrastructure workflows.
• Revamped front-end applications with interactive drag-and-drop interfaces and user-management workflows, increasing engagement by 25% and retention by 15%.

• Built ReactJS and Spring Boot category pages for Hopscotch.in, improving page load speed by 45%.
• Migrated Angular applications to ReactJS, reducing development time by 40% and lowering defect rates across front-end releases.
• Integrated customer-facing features including pinch-to-zoom product images and WeChat API shipment tracking.
• Developed Spring Boot REST APIs for Knowledge Podium, including automated email notifications and a secure user-management dashboard with password hashing and multi-factor authentication, reducing administrative workload by 60%.

Agentic RAG Assistant

• Built a multi-agent assistant in Google ADK with tool-calling agents that plan, route, and orchestrate queries across relational and document stores to answer complex natural-language questions.
• Implemented hybrid retrieval using vector search and SQL, deployed on GCP with tracing, automated evaluation, LLMOps instrumentation, and inference cost monitoring.

Cortex

• Designed a supervisor–worker multi-agent system in Google ADK and LangGraph where specialized agents discover tools, invoke MCP servers, delegate tasks, and coordinate multi-step workflows.
• Built an agentic Graph-RAG pipeline combining vector search and Neo4j, using query decomposition, re-ranking, and reflection loops to produce grounded, citation-backed answers.
• Added guardrails, automated evaluation, distributed tracing, and inference cost optimization for containerized GCP deployments.