Hands-on Capabilities

I don't just manage AI delivery — I build the systems.

Every capability below is backed by public code, a shipped client system, or structured training with practitioners at Google, IIM Visakhapatnam, and SkillfyMe. Repos are open — read the code.

github.com/VinodPungle

Skill matrix

Four stacks that make AI systems real

GenAI & Retrieval

Grounded generation for domains where wrong answers are expensive.

RAG architectureGraphRAGHybrid retrieval Vector databasesNeo4j / knowledge graphsEmbeddings Prompt & context engineeringAzure OpenAIRAG evaluation

Agentic Engineering

From single tool-using agents to orchestrated multi-agent systems.

CrewAI (agents & flows)Multi-agent orchestration Azure Durable Functions agentsAgent tools & APIs Agent memory & sessionsGuardrails & agent securityAgent evals

MLOps & DevOps

The pipelines that keep AI systems reproducible and observable.

Azure DevOpsLLMOpsDVCDagsHub Apache AirflowDockerTerraform / HCP GitHub ActionsMonitoring

Languages & Data

Two decades of Microsoft stack, retooled around Python for AI.

PythonC# / .NETASP.NET MVC / Web API MS SQL Serverpandas · NumPyscikit-learnMatplotlib

Selected repositories

Code you can read today

Working systems, not toy notebooks — retrieval, agents, and the ops scaffolding around them.

hybrid-rag-capstone

Python · Docker · flagship project

Production-grade hybrid RAG — vector + knowledge-graph retrieval, end to end

A complete RAG platform built solo, mirroring the architecture I deliver for enterprise clients: documents flow through an ingestion pipeline into both a vector store and a graph database, retrieval fuses semantic similarity with graph relationships, and every layer is engineered like a product — not a demo.

Modular architecture: ingestion · embeddings · vector_db · graph_db · retrieval · llm · api · ui
Dedicated evaluation and monitoring modules — retrieval quality is measured, not assumed.
Ops from day one: Airflow DAGs for orchestration, DVC for data versioning, Docker Compose deployment, pytest suite, Streamlit demo UI.

RAGGraphRAGAirflowDVCDockerStreamlitpytest

github.com/VinodPungle/hybrid-rag-capstone

Agentic AI

ai-portfolio

CrewAI multi-agent project — defining agent roles, tasks, and crews that collaborate on a goal. The building blocks of agentic systems, in working Python.

CrewAIPython

View repo

Agentic AI

ai-projects

CrewAI Flows experiments, including a guide-creator flow — event-driven agent pipelines where structured steps and agent autonomy are deliberately balanced.

CrewAI FlowsPython

View repo

Agentic CLI tooling

agy-cli-projects

Projects built with agentic CLI coding workflows — including a BigQuery release-notes application — exploring natural-language-first development end to end.

Agentic codingPython · JS

View repo

Infrastructure as Code

hcp-terraform-demo

HCP Terraform workflows — provisioning cloud infrastructure declaratively, the foundation under any production AI deployment.

TerraformHCL

View repo

CI/CD

github_automation_demo

GitHub Actions automation with Docker — pipeline design for continuous integration and delivery of containerised workloads.

GitHub ActionsDocker

View repo

MLOps toolchain

dvcdemorepo · dagshubdemo

Data and experiment versioning with DVC and DagsHub — the reproducibility discipline that separates ML experiments from ML products.

DVCDagsHubJupyter

View repos

How I keep current

Trained where the field is being defined

Google's AI Agents Intensive, IIM Visakhapatnam's executive AI/ML programme, SkillfyMe's MLOps & LLMOps certification, and PMI's GenAI credentials — each described by what it made me build on the Certifications & Trainings page.

Scrum Alliance · 2017

GenAI ×2

PMI GenAI for PMs · 2024

MLOps

SkillfyMe · 2026

IIM-V

Executive AI/ML alumnus · 2025

Explore Certifications & Trainings

Need someone who can review the architecture and run the programme?

That intersection is exactly where I work.

Start a conversation