Ajay Desai

About

Built an AI-native legal research platform that automated analysis across 100,000+ dense legal documents, combining LLM workflows, async backend architecture, and conversational retrieval in production. Also brings cross-domain experience in investment-analysis agents and healthcare claims/billing systems, with a strong emphasis on reliability, deterministic orchestration, and safe handling of messy operational data.

Key Strengths

• Shipped production LLM systems for legal research at scale
• Automated analysis across 100,000+ legal documents
• Built reliable async document pipelines processing 50+ PDFs/hour
• Strong production-readiness mindset for AI systems
• Designs graceful degradation and failure isolation into workflows
• Creates user-friendly AI products with real-time status visibility
• Architects deterministic multi-step agent workflows
• Uses application-controlled tool orchestration for predictability
• Strong evaluation and monitoring approach for agent reliability
• Improved LLM output consistency through context filtering and summarization
• Experienced with messy healthcare/claims data under HIPAA-aligned controls
• Designs safe, idempotent operational workflows to prevent bad downstream actions
• Scaled backend API handling 1M+ trades/day on AWS EKS
• Major performance win by shifting analytics queries from PostgreSQL to Elasticsearch (hours to <30 min; 40% DB load reduction)
• Data consistency strategy for system migration (versioning/timestamps, parallel run validation to 100% match)
• Shipped LLM-powered legal research ingestion/summarization with semantic search (LangChain/OpenAI/Weaviate) end-to-end
• Reliable workflow/agent-style pipeline design with resumable steps, retries/backoff, and manual review escalation (80% fewer failed runs)
• Strong production Python engineering practices (typing/mypy, async, DI, testing, linting, observability with Prometheus)
• PostgreSQL performance tuning (composite indexes, monthly partitioning, window functions) cutting queries 12s to <3s
• Designed scalable async FastAPI microservices for document ingestion and AI orchestration
• Database performance optimization via indexing and batching (reduced nightly batch from 4+ hours to under 2 hours)
• Production RAG architecture integrating OpenAI + LangChain with a vector DB for grounded legal research answers
• Multi-tenant data isolation patterns for healthcare systems with HIPAA-driven security (RBAC + encryption)
• Built Kafka-based streaming pipeline handling 10,000–12,000 records/min with near-instant vital sign updates

Experience

Education

Skills

Languages