Anmol Garg
Open to Data & AI Engineer roles

I engineer data + AI systems that scale, reason, and deliver measurable impact.

3+ years building scalable data platforms and AI-powered systems using LLMs, RAG architectures, and modern data stacks.

Batch→Streaming→RAG→Batch
10M+ records/dayProduction AI pipelinesReal-time + batch systems
Explore My WorkDownload ResumeLet's Build Something Together
System SnapshotProduction-minded
Processed
10M+ records/day
Cloud
AWS + Azure
Pipelines
Batch + Streaming
AI Stack
LLMs, RAG, Vector Search
Records Processed Daily
10M+
Documents Indexed (RAG)
50K+
Retrieval Latency
<300ms
Production Experience
3+ Years
Skills

A structured stack across data engineering, AI systems, and cloud delivery.

Grouped to stay high-signal and scannable for recruiters, while still showing the breadth of systems I can build.

Data Engineering

SparkKafkaAirflowSQL

AI Systems

LangChainLangGraphRAGVector DBs

Cloud & Infra

AWSAzureKubernetes

Programming

Python
AI / GenAI

AI Systems & GenAI Expertise

A practical view of the AI systems work I can build: retrieval pipelines, LLM orchestration, vector search, and the data layers that make those systems reliable.

RAG Systems

Retrieval-Augmented Generation pipelines
Chunking, embeddings, and retrieval optimization

LLM Orchestration

LangChain and LangGraph workflows
Prompt engineering and multi-step execution chains

Vector Databases

Pinecone, FAISS, and Weaviate patterns
Semantic search and context retrieval systems

AI Pipelines

Data to embedding to retrieval to generation flows
Hybrid batch and real-time AI system design

Model Integration

OpenAI and LLM API integration
Fine-tuning awareness and practical production constraints
Architecture Block

Data to intelligence, with clean system boundaries.

A minimal view of how I think about AI system design: reliable data in, structured retrieval in the middle, model output at the edge.

Data
Embeddings
Vector DB
LLM
Response
Live AI Demo

Try My AI System

A lightweight interactive preview that signals applied AI capability, not just portfolio claims.

Suggested Prompts

Chat Preview

Simulated AI response

How would you design a RAG system for 50K+ documents?
I’d separate ingestion, chunking, embedding generation, vector indexing, retrieval, and response composition into clean layers. That makes the system easier to scale, measure, and optimize. Retrieval-first design helps keep answers grounded, while caching and async retrieval keep latency low enough for real user workflows.
System Design

How I design data + AI systems

This section is here to signal engineering judgment: how I think about boundaries, retrieval, observability, and the tradeoffs that matter in production.

01

Separate ingestion, processing, and inference layers so each can scale and fail independently.

02

Use retrieval-first RAG architectures instead of naive prompting when correctness and grounding matter.

03

Design for observability from day one with logs, metrics, and traceable decision points.

04

Optimize for latency, cost, and scalability together rather than overfitting to a single metric.

05

Use hybrid architectures that combine batch, streaming, and inference instead of forcing one pattern everywhere.

Projects

Projects that show how I combine data engineering with intelligent system design.

These are the kinds of systems I want the portfolio to communicate clearly: production-minded data platforms with practical AI capability layered on top.

RAG-Based AI Assistant

Problem

Internal knowledge was scattered across documents and systems, making it slow for users to find the right answer or context.

Architecture

Documents were chunked, embedded, indexed into a vector store, and retrieved into an LLM-driven answer pipeline for context-aware responses.

Tech Stack

LangChainLangGraphOpenAI EmbeddingsPinecone / FAISSPython

Scale

50K+ indexed knowledge chunks across structured and unstructured sources

Impact

Reduced manual search effort and sped up insight retrieval with context-aware responses and a retrieval layer designed for practical trust.

Challenges & Solutions

Challenge: Naive prompting produced shallow answers and no retrieval grounding, making the system unreliable for business-facing usage.
Solution: Built a retrieval-first architecture with chunked documents, embeddings, semantic search, and controlled orchestration for context-aware answers.
Live DemoGitHub

AI Data Pipeline

Problem

AI applications needed a reliable path from raw data to embeddings, retrieval-ready storage, and downstream model interaction.

Architecture

A structured flow handled ingestion, preprocessing, embedding generation, storage, and retrieval, with clear boundaries between data prep and model-serving layers.

Tech Stack

PythonSparkEmbeddingsVector DBBatch + Real-Time Processing

Scale

Hybrid pipeline pattern designed for high-volume ingestion and retrieval-ready indexing

Impact

Created a reusable architecture for converting raw datasets into AI-ready retrieval pipelines with cleaner operational boundaries.

Challenges & Solutions

Challenge: AI applications often break because data preparation, embedding generation, and retrieval layers are loosely connected and hard to operationalize.
Solution: Designed an end-to-end pipeline with deterministic ingestion, preprocessing, embedding storage, and retrieval surfaces that can be monitored independently.
Live DemoGitHub

Streaming + AI Inference

Problem

Real-time systems often stop at ingestion, leaving intelligence delayed when use cases require immediate scoring or anomaly detection.

Architecture

Kafka ingested live events, stream processing prepared inference features, and downstream model logic produced real-time decisions for operational use cases.

Tech Stack

KafkaStreaming PipelinesPythonInference APIsAnomaly Detection

Scale

Real-time event flow designed for high-throughput operational decision systems

Impact

Turned streaming infrastructure into an intelligence layer that supports practical anomaly detection and decision-ready inference.

Challenges & Solutions

Challenge: Streaming systems often stop at transport, while AI use cases need low-latency feature preparation and inference readiness.
Solution: Extended the stream path into an intelligence layer by structuring feature preparation, low-latency scoring, and downstream action hooks.
Live DemoGitHub
About

From data platforms to intelligent systems.

I build systems that connect modern data engineering with practical AI. That means scalable pipelines, solid cloud foundations, and AI workflows that turn retrieval, embeddings, and LLM orchestration into production-ready capabilities.

My background started in data engineering, but the systems I design today are broader and more intelligent: ingestion, transformation, streaming, AI retrieval layers, vector search, and LLM-backed workflows built for real operational use.

Data Engineering
AI Systems
RAG Workflows

I don't just build pipelines. I build systems that make data useful to both people and AI applications.

My engineering depth spans real-time streams, cloud platforms, warehouse design, and practical GenAI system architecture.

I focus on production-ready AI use cases with clear retrieval, orchestration, and reliability boundaries.

Experience

Experience focused on platform reliability, performance, and delivery.

A concise timeline of the environments where I’ve built pipelines, improved data operations, and shipped production-facing systems.

EXL Analytics [UK]

Data Engineer

March 2025 - Present
  • Migrated legacy on-premise processes to cloud-native Spark workflows, reducing processing time by 20%.
  • Built a FastAPI-based microservice on Kubernetes to publish events into Kafka for real-time downstream processing.
  • Partnered with stakeholders across SIT and UAT to reduce release friction and stabilize production delivery.

Reliance Jio

Data Engineer

December 2023 - March 2025
  • Engineered a Hive-driven alerting system for JioFiber outages, reducing downtime resolution time by 40%.
  • Optimized Spark performance to improve batch runtimes by 20% and improve storage efficiency by 15 TB.
  • Built validation pipelines for 50,000+ monthly connections and reduced manual intervention by 80%.

Hexaware Technologies

Data Analyst, Intern

March 2023 - August 2023
  • Developed a Spark-based ETL workflow to automate loading server data into SQL.
  • Replaced a manual Excel-based process with a reproducible pipeline.
  • Improved turnaround time for KPI analysis across stakeholder teams.
Contact

If you're hiring for data + AI systems, let's connect.

Built for teams that need someone who can own reliable data foundations and translate them into production-grade AI systems.

Email

anmolgarg848@gmail.com

LinkedIn

LinkedIn

GitHub

GitHub

Let's Talk

Recruiter-friendly contact