I build AI systems and data pipelines that work on real infrastructure.

My focus sits at the intersection of agent-based AI, data engineering, and cloud-native MLOps — I can wire a LangGraph multi-agent pipeline one day and tune an Isolation Forest on live Prometheus metrics the next.

• 🧠 Current builds: Multi-agent BI systems, Kubernetes ops intelligence, ML-based data quality monitoring
• 📊 Data stack: PostgreSQL · Airflow · dbt · Power BI · Streamlit · Plotly
• ☁️ Infra stack: Kubernetes · Prometheus · Docker · Terraform · Helm
• 🤖 AI stack: LangGraph · LangChain · Groq · RAG · Isolation Forest · scikit-learn
• 📝 Published: YOLOv5 + Raspberry Pi assistive system — IJIRT 2025
• 🎓 Microsoft Learn Student Ambassador

🔭 Dissertation: ML-Based Data Quality Monitoring Framework for Cloud-Native ETL Pipelines

Autonomous 4-agent LangGraph pipeline answering natural language business questions over a proprietary internship dataset

• Built LangGraph StateGraph with 4 typed agent nodes: Planner → SQL Agent → Viz Agent → Insight Agent
• SQL Agent uses few-shot prompting over proprietary schema — 270 scraped Internshala listings, 1,339 skill records
• Auto-selects chart type (bar/pie/line) based on query semantics; generates 2-3 sentence analyst insight per query
• Deployed on Streamlit Cloud with Neon serverless PostgreSQL as backend

LangGraph LangChain Groq LLaMA 3.3 PostgreSQL Streamlit Plotly Neon

ML-driven SRE assistant — detects anomalous pods, retrieves runbooks, generates root cause analysis

• Isolation Forest on live Prometheus time-series (CPU, memory, restarts) — flags statistical outliers across all pods
• TF-IDF RAG retrieves relevant SRE runbooks per anomaly; no heavy embedding models required
• LLM reasoning layer produces structured Root Cause → Evidence → Remediation per flagged pod
• Stress-tested with real fault injection: cpu-stress pod flagged at 1.34 cores → LLM diagnosed "runaway process"

Isolation Forest Prometheus Kind TF-IDF RAG Groq LLaMA 3.3 Helm Streamlit

End-to-end scraping → PostgreSQL → SQL analytics → Power BI dashboard

• Scraped 757 Internshala listings via BeautifulSoup; cleaned to 270 relevant records with 1,339 skill tags
• SQL analysis with CTEs and window functions — stipend benchmarking, skill demand ranking, location heatmaps
• 4-page Power BI dashboard — key finding: ML roles pay 2.4x more than DA roles (₹15.9K vs ₹6.5K/month)

Python BeautifulSoup PostgreSQL SQL Power BI Pandas

GitHub API → SQLite → Streamlit — real DORA metrics across 5 open-source engineering orgs

• Ingested 1,000 PRs, 250 releases, 1,000 issues via PyGithub across dbt-core, ArgoCD, Grafana, Airflow, Prometheus
• Computed weighted health scores: dbt-core 81.2 · ArgoCD 75.0 · Grafana 75.0
• Deployed on Streamlit Cloud with automated data refresh

Python PyGithub SQLite Streamlit Plotly DORA Metrics

886-listing dataset → zone-mapped analytics → Streamlit dashboard

• Mapped 127 localities to 6 Bangalore zones; engineered price-per-sqft metric across segments
• Key finding: Ramamurthy Nagar/K R Puram best value at ₹13-14/sqft vs Whitefield ₹31/sqft

Python Pandas PostgreSQL Streamlit Plotly

Production-grade GitOps pipeline with ArgoCD, Prometheus, Grafana, Terraform

• Automated Kubernetes cluster sync via ArgoCD — Git as single source of truth
• Custom Prometheus alert rules + Grafana dashboards for pod-level observability
• Infrastructure provisioned via Terraform; anomaly detection within 60 seconds of onset

ArgoCD Kubernetes Prometheus Grafana Terraform Helm Flask

AI & Agents

Data Engineering

Infrastructure & MLOps

📝 Published: YOLOv5 + Raspberry Pi Assistive Navigation System — IJIRT 2025
🎓 Microsoft Learn Student Ambassador
📍 Bangalore · Available immediately · On-site / Hybrid / Remote