I'm a Software Development Engineer II at Amazon with 10+ years of experience across backend engineering, cloud-native infrastructure, and developer automation. I enjoy building reliable systems, understanding how they fail, and turning what I learn into practical tools and clear documentation.

You can find my writing, projects, and interactive experiments at jyates.dev.

A tiny language model written in pure TypeScript: trained from scratch, dependency-free, and running entirely in the browser as an interactive terminal dream sequence.

An educational multi-agent orchestration lab built around Google ADK, A2A, and MCP, with agent discovery, reviewer-critic workflows, and distributed tracing.

An OpenAI-compatible embeddings service that runs all-MiniLM-L6-v2 on the AMD XDNA NPU in Strix Halo hardware.

Production-minded starter repositories for Go, Rust, Python, TypeScript, C, C++, Java, and Swift, each with testing, security scanning, CLI examples, and container support appropriate to its ecosystem.

jyates.dev is a serverless portfolio, MDX blog, and browser playground on AWS. Alongside the usual career, project, library, and contact pages, it includes likes and comments, confirmed content subscriptions, RSS, a visitor map, and a browser-only jsh terminal. The interactive tools run in independent draggable windows, including a few deliberately nostalgic under-construction experiments.

flowchart TB
    Browser[Browser]

    subgraph Edge["Edge"]
        DNS[Route 53]
        CF[CloudFront CDN<br/>TLS and security headers]
        CFF[CloudFront Function<br/>SPA and subdomain rewrites]
    end

    subgraph Origins["Origins"]
        Site[(S3 static site<br/>React SPA and prerendered HTML)]
        API[API Gateway REST<br/>origin key, throttle, and quota]
    end

    subgraph Compute["Go Lambda compute"]
        HTTP[HTTP functions<br/>interactions, contact, admin, authorizer]
        Notifications[Notifications<br/>content delivery]
    end

    subgraph State["State and messaging"]
        DDB[(DynamoDB<br/>state, confirmations, checkpoints)]
        SES[SES v2<br/>email and topic contacts]
        Failures[(SQS delivery failures)]
    end

    Browser --> DNS --> CF
    CF --> CFF --> Site
    CF -->|/api/*| API
    API --> HTTP
    Site -->|trusted publish manifest| Notifications

    HTTP --> DDB
    HTTP --> SES
    Notifications --> DDB
    Notifications --> SES
    Notifications -. exhausted retries .-> Failures

    RUM[CloudWatch RUM<br/>scoped Cognito role and budget guard]
    Delivery[Private E2E gate and GitHub Actions<br/>OIDC, Terraform, versioned artifacts]

    Browser -. telemetry .-> RUM
    Delivery --> Site
    Delivery --> HTTP
    Delivery --> Notifications

A few design decisions behind it:

• Static-first React — React 19 and React Router 8 prerender known routes for S3 while preserving client-side navigation. The build also generates RSS, sitemap, robots, and local PlantUML assets from MDX content.
• Purpose-scoped identity — anonymous browser visitor IDs deduplicate likes without using IP addresses as account identifiers. Coarse edge geography powers aggregated visitor statistics.
• Confirmed updates — readers explicitly choose blog and project topics, confirm through a single-use email link, and receive notifications only after a deployment is verified. DynamoDB checkpoints make delivery resumable.
• Release gates before mutation — frontend and API deploy workflows test exact frontend, API, and integration revisions together in LocalStack before assuming an AWS role through GitHub OIDC. No static cloud credentials are used.
• Cost-aware observability — RUM, encrypted logs, X-Ray, dashboards, throttles, and application write limits provide useful production signals while a budget action can disable RUM ingestion at its monthly limit.

The system is split across three public implementation repositories and two private operational repositories:

I write about the useful parts of the work, including the wrong turns. Current areas of exploration include distributed systems and agent protocols, local AI acceleration, production security and cost trade-offs, and low-level hardware investigation. Recent notes and deep dives live on the jyates.dev blog.

The best places to reach me are the contact form or LinkedIn. Project-specific questions and contributions are welcome in the relevant repository.

"Spot the bottleneck, learn fast, ship the fix."