AnishGhule

RISA·Palo Alto, CAfull-time

Founding AI Engineer architecting intelligent healthcare automation systems - from LLM orchestration and AI agent frameworks to FHIR-compliant data pipelines.

• Built AI Agent Factory for vendor-agnostic LLM orchestration across GPT, Claude, and Gemini using Abstract Factory and Adapter patterns
• Pioneered Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols for dynamic tool discovery
• Built PDF Parser Agent and NAR Grid automation achieving >15% zero-touch clinical prior authorization processing
• Established HPAs for 9 microservices; resolved Kubernetes zombie process issues
• Designed FHIR R4 Transformers with Hexagonal/Adapter patterns and Pydantic validation
• Reduced order ingestion from 75 to 7 minutes (10x improvement); saved 500-1000+ hours/year

Sprinklr·Gurugram, Indiainternship

Infrastructure resiliency and observability for enterprise-scale Kubernetes deployments.

• Implemented resiliency checks for CPU underutilization/overusage in Kubernetes; identified and did RCA of issues in 60%+ deployments
• Developed proactive jobs for Elasticsearch shard mapping and fixed replica configuration bugs
• Analyzed and fixed 15+ infrastructure issues using jstack, heap dumps, Grafana, Kibana

Indian Institute of Science (IISc)·Bangalore, Indiaresearch

Micro-expression emotion recognition research under Dr. Punit Rathore.

• Designed architecture using Vision Transformers and Masked Auto Encoders on custom multi-angle facial dataset
• Achieved +11% accuracy improvement on SAMM and CASME2 datasets, outperforming state-of-the-art

Sprinklr·Gurugram, Indiainternship

Backend development and cloud cost optimization for enterprise platform.

• Deployed restricted APIs via SpringBoot; reduced Elasticsearch storage costs ~30%
• Developed Azure Cost Analyzer with 12 drill-down views
• Created Kibana dashboards for forecasted time-series cost visualization

Soroco·Bangalore, Indiainternship

Process mining and intelligent automation using NLP and LLMs.

• Devised rule-engine to summarize action events (4-10 actions to single line, <25% data size)
• Generated natural language descriptions using heuristics and GPT-3
• Anonymized sensitive information via domain-based word mapping

CSIR-CEERI·Chennai, Indiaresearch

Remote photoplethysmography (rPPG) research under Dr. Madan Lakshmanan.

• Implemented Eulerian Video Magnification for remote heart rate estimation
• Generated 20x magnified videos from UBFC-RPPG and proprietary datasets