Generative AI Engineer at Amazon
Shrey
Jindal
Making machines do the boring stuff.
I'd rather build a tool than do the work myself. That philosophy has saved Amazon $30M and kept me entertained on weekends building things nobody asked for.
I don't believe in AGI. I believe in dedicated, improving harnesses — tools that solve specific problems and get better over time. Less hype. More shipping.
Things I've Built
If it spreads without promotion, it works.
UADP — Unified AI Development Pipeline
The magnum opus. A standardized pipeline that transforms task descriptions into validated, ready-to-merge code reviews — no human in the loop until review time.
The insight: AI coding engines are becoming commoditized. The differentiation isn't the model — it's the scaffolding. Constraints, closed feedback loops, and accumulated knowledge that make AI effective in a specific environment. UADP is that scaffolding for Amazon.
Submit a task. Walk away. Return to working code.
SDE 2 — Generative AI
Building AI development infrastructure for Amazon engineers. Three projects, one philosophy: let machines do the boring stuff.
Unified AI Development Pipeline — the magnum opus. Merges context gathering, autonomous coding, and validation into a single fire-and-forget pipeline. Submit a task, return to a working CR.
Amazon's first fire-and-forget coding solution. Describe a task, walk away, come back to a complete code review. No supervision required.
Deep research tool that investigates any ticket across all Amazon internal artifacts — code, wikis, docs, people. Just a kiro-cli harness. Spread entirely by word of mouth.
Software Development Engineer
Full-stack on Amazon's CRM platform. 150,000+ daily agents, millions of customer interactions.
Inherited a publicly exposed static asset vulnerability affecting 150,000+ daily users across global regions. Architected JWT-based authentication at CDN edge with Login PWA architecture — 99.8% attack surface reduction with zero service disruption. Executive recognition.
Real-time AI assistant for abuse detection. Built GraphQL Subscriptions pipeline ingesting live chat transcripts, providing contextual escalation guidance to support reps. Established the foundation for AI-assisted customer service at Amazon.
ML Research Assistant
Enhanced UNet and ResNet-based semantic segmentation models for ecological analysis — specifically satellite imagery classification for environmental monitoring. Improved model accuracy through architecture modifications and training optimizations. Work published in Environmental Research Letters (2024).
And the Rest
The stuff too small or too weird for a portfolio. Open source bits, abandoned experiments, things that worked but weren't worth writing about.
Browse on GitHubCompetition Record
I like building under pressure.
Amazon WWCS 2024
LLM system for CI/CD build log analysis. Automated root cause detection.
Vibehack '25 — SF
Agentic file sorting. Point it at a messy drive, watch it organize.
AI Agent Hack Night 2025
GraphRAG personal assistant. Knowledge graphs meet conversation.
CV × Meta LlamaCon 4
Auto-generated repo documentation. Point and generate.
Odyssey of the Mind 2012
Won India. Competed at World Finals, Iowa State. I was 12.
GrizzHacks 2019
Visual-first interface for food delivery apps. Rethinking how you browse menus.
Michigan State
B.S. Computer Science
December 2022
Competitive — poker, chess, any sport you want to lose at.
Tinkerer — built a telegram memecoin bot and a voice trivia game last month just because.
Watcher — Bollywood movies, F1 races, cricket matches.
Outdoors — hiking when it's warm, snowboarding when it's not.
Ambitious and driven, but honestly? Sometimes a little lazy.
I'd rather build a tool that does the work than do the work myself.
Let's Talk
Building something interesting? I'm always down to chat.