Kirtan Pathak
Data Engineer Ii (IT Analyst Ii) @State of Utah
Signup · Get unlimited contacts
WORK HISTORY
Data Engineer Ii (IT Analyst Ii) @State of Utah
Salt Lake City, UT, US
Architected multi-cloud data infrastructure processing 1TB+ daily across Airflow, AWS DMS, and Oracle RDS — delivering 99% uptime and 30% faster data availability to 5+ government agencies while maintaining zero data loss • Orchestrated 1,100+ production Airflow DAGs integrating SQL Server, Aurora Postgres, and Snowflake; engineered CI/CD pipelines via GitHub Actions reducing production incidents by 80% and eliminating manual deployment risk • Built AI-assisted monitoring infrastructure using GitHub Copilot and Gemini — SLA violation alerting, automated 90-day data retention policies, and end-to-end load monitoring across tables in DEV and PROD; accelerated complex scheduling logic cutting development cycles while maintaining zero data loss standards • Spearheaded migration of 700+ legacy Oracle procedures to Snowflake — 40% query performance gains,$200K+ annual cost savings, and advanced SQL tuning improving system throughput by 25% supporting 2x user growth • Implemented real-time streaming via Airbyte and AWS reducing executive reporting latency by 35%; designed SFTP-to-S3 file processing pipelines with DuckDB-based Python ETL and Airflow-native Oracle integration • Designed comprehensive alerting system tracking SLA violations, performance anomalies, and execution failures across mission-critical workflows — reducing MTTR by 60% across enterprise data infrastructure
EDUCATION
CHARUSAT
Bachelor of Technology, Computer Science and Engineering
The University of Texas at Dallas
Master of Science - MS, Computer and Information Sciences, General
The Bright School
High School, Gujarat Board
ABOUT KIRTAN PATHAK
I don\'t just move data — I make it intelligent.By day, I architect enterprise data infrastructure at the State of Utah, processing 1TB+ daily across multi-cloud environments. I\'ve built and orchestrated 1,100+ production Airflow DAGs with zero data loss, migrated 700+ legacy Oracle procedures to Snowflake unlocking 40% query performance gains and $200K+ in annual cost savings, and engineered CI/CD pipelines that cut production incidents by 80%. When something breaks at 2am in a government data system, I\'m the person who built it not to break.That foundation — knowing what production really means — is exactly why I build AI systems differently.NeuralVault — AI Learning MemoryMost RAG tutorials show you how to call an API and get an answer. I wanted to build something that actually works under pressure:→ Hybrid retrieval (BM25 + vector semantic search combined)→ Cross-encoder reranker (ms-marco-MiniLM) for precision scoring→ ChromaDB vector store with local MiniLM embeddings→ Groq-hosted Llama 3.3 70B for grounded answer generation→ Citation enforcement — refuses to hallucinate, flags insufficient evidence→ RAGAS evaluation pipeline with CI quality gatesSLM-Bench-Router — Local SLM Benchmark PlatformMost teams pick one model and hope it works for everything. I wanted to prove that assumption wrong with data. So I benchmarked 3 local small language models (llama3.2, phi4-mini, qwen2.5:3b) across 3 structured tasks with 2,250 grammar-constrained inference calls — then used the findings to build a RouterAgent that dispatches every request to the empirically proven best model:→ qwen2.5:3b for log classification (96% accuracy, 9.5 tok/s) and document extraction (100%)→ llama3.2 for code review (100% accuracy, 8.8 tok/s — 2x faster than qwen on same task)→ phi4-mini as router (94% classification accuracy) — a result that only emerged from benchmarking, not intuition→ Pydantic v2 schema enforcement with error-feedback retry loops across all agents→ 100% local via Ollama — zero cloud dependencies, zero API costsTwo projects. Two different problems. Same principle: build it so it actually works, measure everything, let the data make the decisions.I\'m looking for AI Engineer or Data & Retrieval Engineer roles where I can bring both sides together — the data foundation and the AI system on top of it.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.