Hi, I'm

Kevin Shah

Software Engineer | AI Engineer

I build and operate production LLM systems — agentic workflows, RAG pipelines, and multi-cloud AI platforms with systematic prompt evaluation, observability, and safety guardrails.

I take AI systems from prototype to production: LLM orchestration platforms, retrieval-augmented pipelines, prompt evaluation/optimization, and the observability and guardrails that keep them reliable at scale. Currently an AI Engineer on the MLOps platform team at DNV; previously a Software Development Engineer at Amazon. Open to AI Engineer, MLOps, and Forward Deployed roles in the USA, Canada, Europe, and remote.

📍 Houston, TX — Open to USA, Canada, Europe & Remote

Get In Touch Resume GitHub

Selected Work

Production LLM systems I designed and shipped at DNV — problem, approach, and measurable outcome.

Eval-Driven Prompt Platform

DNV

Turning prompt engineering from guesswork into a measured pipeline.

LLM Wiki — Knowledge Synthesis

DNV

Making AI coding assistants faster and cheaper by replacing raw source with synthesized wikis.

Agent Long-Term Memory (RAG)

DNV

Personalized, context-aware agents backed by vector retrieval.

Case study details are access-controlled

Full problem, approach, and outcome are available on request. If you have an access code, enter it below.

Don't have the code? Reach out via the contact form.

Experience

My professional journey building LLM systems, ML platforms, and backend services.

AI Engineer, Machine Learning Operations

Current

DNV·May 2023 – Present·Houston, TX

▸Designed and own a production LLM orchestration platform for agentic workflows with persistent state, multi-step reasoning, and fault tolerance — built with Python, FastAPI, LangChain, and LangGraph.
▸Built a prompt-management system with an LLM-based optimizer for systematic prompt versioning, evaluation, and refinement — cut a 20-hour manual task to 3–4 hours (~85%) while reducing hallucinations and improving output reliability.
▸Created an "LLM Wiki" knowledge-synthesis layer that feeds AI coding tools (Kiro, Claude) synthesized repo context instead of raw source — ~60–70% lower token usage and ~80% faster AI responses.
▸Built a long-term agent memory system on MongoDB vector embeddings, retrieving top-k semantically similar memories per query for personalized, context-aware responses at scale.
▸Deployed AWS Bedrock Guardrails across environments via CloudFormation IaC — content safety, PII filtering, and hallucination mitigation at the inference layer.
▸Implemented end-to-end LLM observability (distributed tracing, token-usage metrics, agent-step dashboards), cutting incident triage time and improving production visibility.
▸Architected serverless backend on AWS Lambda, ECS, and S3 (~50% lower infra cost) and a state-machine-inspired workflow engine integrating SageMaker (~40% lower execution latency).
▸Led backend design and architecture reviews, standardized build/deploy workflows, and mentored junior engineers and interns.

PythonFastAPILangChainLangGraphAWS BedrockSageMakerAWS LambdaECSMongoDBRedisCloudFormationVue.js

Software Development Engineer

Amazon.com Services LLC·Jun 2022 – Mar 2023·Austin, TX

▸Led backend changes supporting customer-experience analysis across 21 global marketplaces, contributing to a zero-downtime migration impacting 550M+ users.
▸Built runtime monitoring and alerting for latency and customer-impact metrics, enabling real-time detection of production issues at global scale.
▸Designed and implemented backend components for a large-scale Order Summary system, integrating with multiple critical services and legacy systems.
▸Participated in on-call rotations, independently diagnosing and resolving high-severity production incidents and contributing to operational reviews for a global team.

JavaPythonSQLShellAWS

Education

Master of Science in Computer Science

Arizona State University·GPA 3.81 / 4.00·May 2022·Tempe, AZ

Bachelor in Computer Engineering

Sardar Vallabhbhai National Institute of Technology (NIT Surat)·Jul 2020·Surat, India

Projects

Things I've shipped outside of work — published, public, and clickable.

sonar-complexity

Published

VS Code extension that surfaces SonarQube cognitive and cyclomatic complexity metrics inline in the editor, so you see hotspots while you code instead of after a CI scan.

TypeScriptVS Code APISonarQube

View live Source

AlgoTrader

In Progress

Backend-focused algorithmic-trading app that backtests and paper-trades quantitative strategies. Ingests historical market data via the Alpaca API, computes technical indicators (MACD, RSI, Bollinger Bands), and simulates trades visualized through a custom React + TradingView interface.

PythonFastAPIReactAlpaca API

neetcode-submissions

My NeetCode.io problem submissions

Python

CardGames

Mobile scorekeeper for Declare, Judgement, and 3 of Spades — built with Expo + TypeScript

card-games-rosy.vercel.app

TypeScript

View all repositories on GitHub

Skills

Technologies and tools I work with day-to-day.

MLOps / AI

LangChainLangGraphMCPLiteLLMRAG PipelinesPrompt Engineering & OptimizationLLM EvalsAWS BedrockBedrock AgentCoreBedrock GuardrailsAgentic WorkflowsEmbeddings & Context Retrieval

Languages & Scripting

PythonJavaScriptTypeScriptSQLBash

Backend & API

FastAPINode.jsRESTful APIs

Cloud & DevOps

AWS LambdaAWS ECSAWS ECRS3SageMakerCloudWatchCloudFormationDockerGit

Databases

MongoDB (Vector)PostgreSQLRedis

Frontend & Reliability

Vue.jsNuxt.jsReactD3.jsUnit & Functional TestingProduction Observability

Get In Touch

Open to AI Engineer, MLOps, and Forward Deployed roles in the USA, Canada, Europe & remote.

Whether you have a role in mind, want to collaborate on a project, or just want to say hi — my inbox is always open. I'll do my best to get back to you promptly.

Email
kevinjshah2207 [at] gmail [dot] com

Location
Houston, TX · Open to USA, Canada, Europe & Remote

GitHub LinkedIn

* Email shown as plain text to reduce spam. Use the form to send a message directly.