Skip to content
View DevMLAI01's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report DevMLAI01

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
DevMLAI01/README.md

Hi, I'm Saurabh Kumar Dewangan 👋

Data & AI Leader | 19 Years Delivering at Enterprise Scale | Driving AI Adoption Across Data, Analytics & Engineering

LinkedIn GitHub Calgary, Canada Profile views


About Me

I'm a Technology Lead with 19 years of enterprise experience in Telecom and Data Engineering, transitioning into senior leadership roles at the intersection of Data Analytics, Data Engineering, and AI adoption.

I've spent nearly two decades owning complex, mission-critical systems — and I now apply that foundation to lead the design and delivery of modern data platforms, AI-powered analytics pipelines, and intelligent agentic systems. I bring both the strategic perspective to align data & AI investments with business outcomes, and the technical depth to architect and validate the solutions my teams build.

  • Driving: Enterprise AI adoption · Data platform modernisation · Agentic AI integration into business workflows
  • Architecture focus: Azure Databricks Lakehouse · Medallion Architecture · Multi-agent LLM systems · RAG pipelines
  • Targeting: Manager | Senior Manager | Architect roles in Data Analytics · Data Engineering · AI in Canada & India
  • Strength: Translating ambiguous business problems into scalable data and AI architectures — and shipping them
  • Based in Calgary, AB, Canada | Open to remote and hybrid roles

Featured Projects

Medallion Architecture on Azure Databricks — anomaly detected to incident report in <60 seconds

Production-style telecom data lake processing ~55K CDRs, 5K KPIs, and 70 unstructured NOC files through a full Bronze to Silver to Gold pipeline. Z-score SQL alert fires a webhook to an Azure Function, which runs a GPT-4o tool-calling loop that queries live Gold tables, searches NOC history, and writes structured Markdown incident reports directly to ADLS — fully automated, end-to-end.

Azure Databricks Delta Lake AutoLoader Unity Catalog Azure OpenAI GPT-4o Azure Functions ADLS Gen2 PySpark


LangGraph + RAG + AWS Lambda — 45-90 min NOC workflow compressed to <60s

4-node self-correcting state machine that autonomously investigates alarms, retrieves vendor SOPs via semantic search, drafts resolution tickets, and safety-audits its own output before returning — deployed on AWS Lambda with a live REST endpoint. 35 tests · 70% coverage · 3-retry self-correction loop.

LangGraph GPT-4o AWS Lambda DynamoDB RAG LangSmith


Multi-agent system targeting EC2 instances: CPU <5% over 7 days AND cost >$100/month

5-node orchestrator that scans AWS, classifies resources via RAG (faithfulness >= 0.85), generates Terraform PRs, and enforces mandatory human approval before any change — zero autonomous terraform apply, full audit trail on every decision. $0 infrastructure cost.

LangGraph Gemini Qdrant MCP Terraform Ragas HITL


Role-based document access · guardrails · cost monitoring · Ragas evaluation pipeline

Production-grade RAG chatbot where retrieval is filtered by user roles — different personas see different document subsets. Integrates guardrails to prevent hallucination and prompt injection, tracks per-query token costs, and ships a Ragas evaluation harness for faithfulness and answer relevancy scoring.

RAG RBAC LangChain Guardrails Ragas ChromaDB Python


LLaMA 3.1-8B fine-tuned with LoRA — 16GB model to ~5GB, only 0.8% of weights trained

Domain-adapted fine-tuning pipeline for Netezza SQL to PySpark translation using 4-bit NF4 quantization and LoRA (rank-16). Runs on a free Colab T4 GPU and exports to GGUF format for production deployment via Ollama or llama.cpp.

LLaMA 3.1 LoRA/PEFT Unsloth bitsandbytes GGUF TRL


3 Claude models · 5 security guards · live on AWS EC2 · self-correcting SQL pipeline

4-node pipeline routing natural language through Claude Haiku (retrieval), Opus (SQL generation with Pydantic guards), SQLAlchemy executor, and Sonnet (business narrative) — with AST-based table allow-listing and automatic self-correction for failed queries.

LangGraph Claude Streamlit SQLAlchemy Pydantic AWS EC2


Tech Stack

AI / ML

LangGraph LangChain PyTorch HuggingFace OpenAI Google Gemini Claude

Data Engineering & Lakehouse

Azure Databricks Delta Lake Apache Iceberg PySpark Apache Airflow Apache Kafka SQL Python

Cloud & Infrastructure

Azure AWS Terraform Azure Functions Docker Kubernetes

MLOps & Evaluation

LangSmith Ragas Qdrant MLflow


GitHub Stats

GitHub Stats Top Languages

GitHub Streak


What I Bring to a Data & AI Leadership Role

Area What I Deliver
Strategic Leadership Define data & AI roadmaps aligned to business goals; translate C-suite priorities into architecture decisions and delivery plans
AI Adoption & Diffusion Drive enterprise-wide AI adoption — from use-case identification and build-vs-buy decisions to team enablement and governance frameworks
Data Platform Architecture Design and deliver cloud-native lakehouses (Azure Databricks, Medallion Architecture, Delta Lake, Unity Catalog) that scale to enterprise data volumes
AI & Analytics Systems Architect production LangGraph agents, RAG pipelines, agentic triage systems, and LLM fine-tuning workflows that solve real operational problems
Cross-functional Delivery 19 years leading engineers, aligning stakeholders, and shipping complex data systems in high-stakes Telecom and cloud environments
Engineering Excellence HITL guardrails, Ragas evaluation, observability (LangSmith/MLflow), PII masking, RBAC — building AI systems that are safe and auditable, not just functional

Let's Connect

I'm actively targeting Manager · Senior Manager · Architect roles in Data Analytics · Data Engineering · AI in Canada and India.

If your organisation is modernising its data platform, scaling analytics capabilities, or figuring out how to responsibly adopt AI at enterprise scale — that is exactly the kind of challenge I want to help solve.

Reach me via LinkedIn


"19 years of building systems that work at scale. Now building systems that think."

Pinned Loading

  1. telecom-noc-agent telecom-noc-agent Public

    LangGraph-based AI agent for Telecom Network Operations Center (NOC) with RAG-powered troubleshooting

    Python 2 1

  2. autonomous-sre-finops autonomous-sre-finops Public

    Autonomous SRE & Cloud FinOps Orchestrator: LangGraph + Gemini + Qdrant + MCP to detect underutilized AWS resources and open Terraform PRs with HITL approval

    Python 1

  3. Legacy-to-Cloud-tuner Legacy-to-Cloud-tuner Public

    Fine-tune LLaMA 3.1-8B to translate legacy Netezza SQL into optimized PySpark DataFrame code — runs on Google Colab Free Tier (T4 GPU)

    Jupyter Notebook 1

  4. text-to-sql_Mult_Agent-bi-orchestrator text-to-sql_Mult_Agent-bi-orchestrator Public

    Python 1