Anton Glenbovitch
Senior AI Engineer • LLM Systems • RAG • AWS Architecture
I build backend systems, RAG workflows, evaluation harnesses, and AI automation patterns for enterprise-style environments. My focus is reliability, testability, cost control, observability, and practical deployment.
20+ Years Building Enterprise Systems
Backend architecture, enterprise applications, data workflows, integrations, and AI-enabled engineering systems. Experience across university IT, health insurance, and independent consulting.
Core Expertise
- AI Systems: RAG workflows, LLM evaluation, prompt/control-plane design, agentic orchestration
- Backend Engineering: Python, FastAPI, Java, REST APIs, SQL, enterprise integrations
- Cloud / DevOps: AWS Lambda, Bedrock, OpenSearch, Docker, CI/CD, observability
- System Design: reliability, testing, guardrails, monitoring, cost-aware architecture
Engineering Philosophy
In production AI systems, the main challenges are not model selection, but:
- Data quality and preparation
- Retrieval accuracy and ranking
- System design and integration
- Evaluation and continuous monitoring
Reliable systems require guardrails, evaluation metrics, and rigorous testing to control hallucinations and maintain consistency at scale.
Featured Projects
Selected engineering projects showing RAG architecture, evaluation workflows, cloud patterns, guardrails, and backend system design.
Liquid in Glass Particle Simulation
Overview: Browser-native fluid experiment that simulates layered liquids inside a glass with real-time particle physics, tilt/shake/swirl interactions, adjustable particle density, dynamic color blending, and an optional 3D depth mode.
Enterprise Claim AI Platform
Overview: Reference architecture for AI-assisted insurance claim analysis using RAG, workflow orchestration, fraud-risk scoring, audit logging, and human review patterns. Designed to demonstrate how an enterprise claims workflow could combine retrieval, LLM reasoning, evaluation, and governance.
Key Technical Decisions
- Hybrid Retrieval: Combined semantic search + BM25 ranking for 8% accuracy improvement over semantic-only approach
- Evaluation Framework: Automated metrics (ROUGE, BERTScore) + human QA labels for ground truth validation
- Cost Optimization: Prompt caching (30% reduction), batch processing, cheaper embedding models
- Guardrails: Hallucination detection, conservative "I don't know" responses for out-of-domain queries
Impact
- Demonstrates how to structure claim-analysis workflows with retrieval grounding, model routing, evaluation checks, human fallback, and auditable decision metadata.
- Shows a practical cloud pattern for regulated AI systems where traceability, conservative responses, and escalation matter as much as model output.
RAG Evaluation Framework
Overview: Comprehensive evaluation framework for assessing RAG pipeline quality. Combines automated metrics with human-in-the-loop validation to measure retrieval accuracy, generation quality, and hallucination rates.
Health Insurance Member Q&A Chatbot
Overview: Full-stack conversational AI system helping health insurance members answer questions about coverage, claims, benefits. Combines React frontend, Node.js backend, and RAG pipeline for accurate, compliant responses.
Recent Articles
Technical deep dives on RAG, LLM systems, and production AI architecture.
Spec-Driven Development: Moving AI Coding from Experimentation to Production Discipline
Why AI coding needs a contract-first workflow with clear specifications, planning gates, traceable implementation, and verification before production use.
Read on Website →Building Production RAG: Cost Optimization Strategies
How to reduce RAG pipeline costs by 66% without sacrificing quality. Covers prompt caching, embedding model selection, batch processing, and cost-per-query optimization.
Read on Medium →Evaluating RAG Systems: Beyond Automated Metrics
Why automated metrics alone fail for RAG evaluation. The case for human-in-the-loop validation, building labeled datasets, and continuous monitoring in production.
Read on Dev.to →Conversational AI in Healthcare: Domain-Specific Challenges
Lessons from building health insurance chatbots. PII handling, regulatory compliance, conservative response strategies, and maintaining accuracy in regulated domains.
Read on Medium →Architecture Lessons from 20 Years: Systems Thinking at Scale
Evolutionary lessons building systems from monoliths to microservices to serverless. Why governance matters, cost is architecture, and observability is a first-class concern.
Read on Dev.to →About
Experience
20+ years building enterprise systems across backend platforms, data systems, and AI applications.
- ITA Consulting / Independent Consulting: AI systems, backend architecture, automation workflows, and enterprise technology strategy.
- Yale University ITS: enterprise application development, backend systems, integrations, SQL/data workflows, production support, and cross-functional delivery.
Core Strengths
- Technical: 20 years systems architecture, Java/Python, AWS, databases, microservices
- AI/LLM: 1+ year production RAG, LangChain, vector DBs, evaluation frameworks
- Domain: 10 years health insurance (claims, compliance, enterprise integration)
- Leadership: Cross-functional team leadership, stakeholder communication, technical strategy
Let's Work Together
Interested in building production AI systems? Have a RAG or LLM architecture question? Let's talk.
Email: a.glenbovitch@gmail.com
LinkedIn: linkedin.com/in/anton-glenbovitch-9934897a
Phone: available upon request