Sai Harsha Kondaveeti
London, UK · Open to AI/LLM Engineering Roles
Building production-grade AI systems with agentic architectures, retrieval-augmented generation, and evaluation-driven ML infrastructure.
What I Build
Production-grade AI systems with measurable outcomes, structured orchestration, and deployment readiness
Agentic Systems
Designing structured multi-agent workflows with clear execution boundaries and controlled orchestration.
Retrieval-Augmented Generation (RAG)
Modular RAG pipelines with explicit separation of retrieval, orchestration, prompting, and evaluation.
Production ML Pipelines
Scalable ML systems with observability, reliability, and deployment readiness.
Featured Projects
View AllRAG Foundry
Framework for building RAG systems with separated retrieval, ranking, and generation components, plus tools to measure quality across datasets.
Core Idea
Separate retrieval, orchestration, prompting, and evaluation into independent, testable modules
Architecture
Plugin-based retriever system with swappable vector stores and configurable chunking strategies
What You Can Reuse
Evaluation harness with dataset-driven metrics (GitHub template available)
Agiorcx Lib
Library that provides an execution layer for AI agents, with explicit control flow, error handling, and guardrails for production use.
Core Idea
Explicit control flow for agents with guardrails, rollback mechanisms, and execution logging
Architecture
State machine-based orchestration with pre/post execution hooks and audit trails
What You Can Reuse
Agent coordinator pattern with guardrails (library + examples)
Evallit
Evaluation toolkit for LLM and RAG systems that runs dataset-based checks, records metrics, and exposes hooks for monitoring and debugging.
Core Idea
Define test datasets upfront, run automated evaluation passes, track metric trends over iterations
Architecture
Pluggable metric system (ROUGE, semantic similarity, custom scorers) with experiment tracking
What You Can Reuse
Evaluation pipeline template with metric collectors and reporting dashboards
LIA Swarm
Proof-of-concept multi-agent system that applies RAG Foundry design principles to coordinate agents across retrieval, reasoning, and response steps.
Core Idea
Demonstrate coordinated multi-agent workflows with shared context and structured communication protocols
Architecture
Message bus for agent communication with priority queuing and context inheritance
What You Can Reuse
Multi-agent coordination pattern with message schemas (reference implementation)
Sai Harsha Logs
View AllNotes & Lessons from AI engineering and system building
Why Most RAG Systems Fail in Production
Common architectural mistakes that cause RAG systems to fail at scale and how to build production-ready alternatives.
If Your AI Agents Never Fail, You're Missing Critical Evaluation
Why perfect agent execution is a red flag and how to design failure-aware evaluation systems.
Stop Measuring AI Systems the Wrong Way
Moving beyond accuracy metrics to build meaningful evaluation frameworks for production AI systems.
Subscribe to Garvamanai
Get weekly insights on AI engineering, system design, and building production-grade ML systems. Join a community of engineers who care about reliability and measurable outcomes.