Lead AI Engineer

Marching towards Singularity...

I build production-ready GenAI platforms that teams can trust in real workflows, not just demos.

My focus is end-to-end delivery: model behavior, data quality, cloud infrastructure, and secure rollout.

I care about practical outcomes: faster engineering loops, safer data handling, and measurable business impact.

Explore experience

Experience

A focused timeline of roles with outcomes.

Jul 2024 — Present
Noida, India

Lead AI Engineer · CoForge Ltd.

GenAI data products for enterprise platforms

  • Led 2 GenAI data products - Synthetic Test Data Generator (STDG) and Automated Data Classifier - delivered to Philip Morris International (Global FMCG enterprise).
  • Reduced test-data preparation effort by 70–80% using LLM-powered synthetic data generation on Snowflake metadata.
  • Tagged 100% of Snowflake columns using LLM-driven metadata intelligence (Snowflake + Atlan), including PII detection and explainable reasoning.
  • Owned AWS infra (ECS, ALB, Bedrock, S3, CloudFront, GuardDuty) with Terraform + CI/CD and security reviews.
SnowflakeAtlanAWS BedrockTerraformECS
Nov 2023 — Apr 2024
Hyderabad, India

Consultant AI & ML · ValueLabs LLP

LLM platforms, fine-tuning & inference at scale

  • Led development of AiDE, an enterprise LLM platform powering code copilots (brownfield projects) and NL→SQL analytics across deployments.
  • Pre-trained and instruction-tuned LLMs (3B–40B) using FSDP/DeepSpeed + PEFT (LoRA/QLoRA); built a copilot exceeding GPT‑3.5 on internal HumanEval.
  • Shipped NL→SQL for Postgres/MySQL, improving query accuracy by 30% and deployed to 3 clients, generating $500K+ yearly savings.
  • Productionized inference with vLLM (TensorRT) and TGI, supporting 250 concurrent users with safety guardrails and monitoring.
DeepSpeedLoRAvLLMTensorRTTGI
Oct 2021 — Oct 2023
Hyderabad, India

Associate Consultant AI & ML · ValueLabs LLP

GenAI, NLP, CV, Predictive Analytics

  • Trained and fine-tuned LLMs (Falcon, CodeLlama, LLaMA2, Pythia, MPT, Flan‑T5) on NVIDIA A100 clusters.
  • Delivered 20+ POCs and 8 MVPs across GenAI/NLP/CV and moved multiple solutions into production.
  • Awarded Super Star of the Quarter for consistent delivery under aggressive timelines.
PyTorchTransformersLLMAzure
Jul 2020 — Sep 2021
Hyderabad, India

Technical Consultant AI & ML · ValueLabs LLP

Document intelligence & API engineering

  • Built OCR + NLP extraction pipelines with 97% accuracy across identity documents.
  • Shipped invoice and legal document extractors using ELECTRA and BERT‑QA, reaching 92% production accuracy.
  • Designed and deployed 20+ REST APIs using Django/FastAPI, containerized with Docker and hosted on AWS EC2.
FastAPIDjangoOCRAWS
Dec 2019 — May 2020
New Delhi, India

ML Research Trainee · DRDO (DTRL)

Hyperspectral imagery research

  • Developed an unsupervised band-selection approach for hyperspectral imagery using multi-objective optimization and Boltzmann entropy.
OptimizationRemote sensingResearch

Selected Work

Synthetic Test Data Generator (STDG)

GenAI · Data Platform

LLM-powered synthetic data generation using Snowflake metadata to replace production data in dev/QA.

  • Reduced test-data preparation effort by 70–80%.
  • Enabled compliant rollout across initial 10 global markets.
SnowflakeLLMsAWS

Automated Data Classifier

Governance · PII

Column-level classification for Snowflake using LLM-driven metadata intelligence with explainable reasoning.

  • Tagged 100% of Snowflake columns (PII detection + exports).
  • Natural-language controls for analysts and data owners.
AtlanSnowflakePIIRAG

AiDE (AI-Driven Engineering)

LLM Platform

Enterprise LLM platform powering code copilots and NL→SQL analytics, optimized for cost and latency.

  • Instruction tuning (3B–40B) using DeepSpeed/FSDP + LoRA/QLoRA.
  • NL→SQL accuracy improved by 30%; deployed to 3 clients with $500K+ yearly savings.
  • High-throughput inference via vLLM (TensorRT) and TGI for 250 concurrent users.
DeepSpeedvLLMTensorRTPostgres

Claims Automation (Smart Virtual Adjuster)

Insurance

End-to-end ML pipeline for claims triage and automation.

  • Trained XGBoost on 400k historical claims; achieved 87% accuracy/recall.
  • Reduced manual workload by 40% via AWS SageMaker pipeline.
AWS SageMakerXGBoostSnowflake

Fraud Monitoring Solution

Anomaly Detection

Procurement anomaly detection across 10 years of contracting data.

  • Flagged $10B worth of tenders for investigation.
  • Azure Databricks pipelines + Power BI dashboards.
Azure DatabricksiForestPower BI

Document Intelligence APIs

NLP · OCR

OCR + extraction systems for identity, invoice, and legal documents with production APIs.

  • Identity docs extraction at 97% accuracy; invoice/legal at 92%.
  • Built 20+ REST APIs using Django/FastAPI, Dockerized on AWS.
FastAPIDjangoELECTRABERT-QA

Blog

Case studies and architecture diagrams — coming soon.

GenAI product case studies

Problem → approach → architecture → results. Stay tuned.

Architecture diagrams

Simple diagrams (RAG, training pipelines, inference stacks) with trade-offs.

Build notes

What worked, what didn’t, and patterns I reuse in production.

Stack

Tools I use to ship production systems.

Backend

Python · FastAPI · Django · SQL

GenAI

LangChain · LangGraph · RAG · MCP · A2A · Agentic AI · AWS Bedrock · Azure AI Studio

Training & Inference

PyTorch · PEFT · DeepSpeed · Flash Attention · vLLM · TGI · TensorRT

Cloud & Data

AWS · Azure · Databricks · Snowflake · Postgres

DevOps & Security

Terraform · Jenkins · Docker · SonarQube · Vault · GuardDuty

Libraries

Pandas · NumPy · scikit-learn · XGBoost · OpenCV · spaCy

Credentials

Education

B.Tech, Computer Science & Engineering
SRM Institute of Science and Technology · 2016 — 2020

Certifications

Microsoft AI‑900
Azure AI Fundamentals
Microsoft DP‑900
Azure Data Fundamentals
Microsoft AZ‑900
Azure Fundamentals
Microsoft SC‑900
Security, Compliance & Identity

Contact

If you’d like to collaborate or chat, feel free to reach out.

Send a message

This form doesn’t store anything — it opens your email client with a pre-filled draft.