~/hero
I build AI systems that survive production, not just demos.
Cloud data engineer, 5+ years — AI agent pipelines, MLOps, and the infrastructure that keeps both running.
- 5+
- years shipping data infra
- 100M+
- records in production
- 60%
- ETL throughput gained
- 45%
- runtime cut on a lakehouse
~/about
About
Saketh Bantu is a cloud data engineer who's spent the last five years building the infrastructure behind commercial analytics — most recently architecting AI agent pipelines and MLOps platforms for one of the largest CPG companies in the US. His path started in BI and analytics at a life-sciences AI startup, moved through IoT and predictive-maintenance infrastructure at Toyota Forklift, and now spans the full stack from data pipelines to production AI systems at Church & Dwight. He holds an MS in Cybersecurity and Information Security from Eastern Illinois University, and published research on deep-learning-based galaxy classification.
~/projects
Featured projects
Grounded support agent
RAG agent that cites sources and admits what it doesn't know.
Incident-triage agent
Reads pipeline logs, classifies failures, drafts a root-cause note.
Self-healing streaming pipeline
Kafka to Delta Lake with automatic data-quality alerts.
Retail KPI dashboard
Interactive BI dashboard on public CPG sales data.
~/experience
Experience
RAG knowledge pipeline
- problem —
- Stakeholders needed faster access to enterprise knowledge scattered across internal sources.
- approach —
- Architected and operationalized a RAG pipeline with LangChain and the OpenAI API.
- result —
- Automated retrieval and report summarization for business stakeholders.
Commercial-scale ETL
- problem —
- Analyzing customer usage patterns across 100M+ records demanded a reliable, high-throughput pipeline.
- approach —
- Built ETL/ELT workflows in PySpark and Azure Data Factory, then optimized the Databricks SparkSQL layer.
- result —
- 60% throughput improvement at 100M+ record scale.
Production MLOps and observability
- problem —
- Models needed reliable, low-latency serving with full production visibility.
- approach —
- Containerized inference services on Docker/Kubernetes; added Prometheus/Grafana observability and Great Expectations validation.
- result —
- 100% data integrity maintained on critical production reports.
Real-time IoT pipeline
- problem —
- Predictive maintenance needed real-time processing of time-series sensor streams.
- approach —
- Designed event-driven ETL with Kafka and PySpark.
- result —
- A scalable real-time pipeline supporting predictive-maintenance models.
Lakehouse and self-healing CI/CD
- problem —
- ETL runtime and deployment reliability needed to improve as the lakehouse grew.
- approach —
- Built an S3 + Delta Lake + Redshift lakehouse with partition pruning and caching; Airflow + Lambda orchestration with automated schema validation on EKS.
- result —
- 45% ETL runtime reduction and a self-healing deployment pipeline.
Governed BI reporting
- problem —
- The business needed reliable KPI dashboards with strict governance over sensitive commercial data.
- approach —
- Built Tableau/Power BI data models; implemented role-based security and authentication on Tableau Server; optimized performance with warm caching.
- result —
- Governed, high-performance reporting adopted for strategic decisions.
~/skills
Skills
Cloud Platforms
Data Lake & Warehousing
Generative AI & MLOps
Data Engineering
Microservices & DevOps
Visualization & Reporting
Languages
~/credentials
Certifications, education & research
Certifications
- Microsoft Certified: Azure Data Fundamentals (DP-900)
- Microsoft Certified: Azure Security Engineer Associate (AZ-500)
- AWS Cloud Practitioner – Trained
Education
MS, Cybersecurity and Information Security
Eastern Illinois University · 2023
BTech, Computer Science and Engineering
JNTUH · 2020
Publication
“Galaxies classification using Deep Learning Algorithm in Convolutional Neural Networks” — UGC Care Group, May 2020
~/contact
Get in touch
Open to cloud data engineering and applied AI/MLOps roles. The fastest way to reach me is email or LinkedIn.