Abhishek Joshi, full-stack developer and AI engineer in Perth, Australia

Hi, I'm Abhishek. I build full-stack products with AI at the core.

16+ years of shipping software in .NET, Angular, TypeScript and Python. Today that means LLM-powered features, AI assistants and the full-stack apps around them, built to hold up in production.

Start a conversation github.com/pixelbyaj

What I build

AI is only useful when it ships inside a real product. That's the part I like most.

AI assistants and copilots

Chat experiences with memory, grounded in your own data and wrapped in guardrails, built into the product rather than bolted on.

Azure OpenAI, .NET, TypeScript

Document and image intelligence

LLMs and vision models that read emails, documents, forms and images, and turn them into structured data your systems can trust.

Azure AI Vision, Microsoft Graph, Python

Full-stack web products

Angular and TypeScript front ends on .NET and Python APIs, designed, built and deployed end to end, from the first sketch to production.

Angular, TypeScript, .NET, Azure

Real-time systems and integration

Event streams and message queues that connect AI features to everything else, including banking and ISO 20022 payment systems.

Kafka, SignalR, IBM MQ

How I build with AI

Sixteen years of shipping software is what lets me tell a good AI suggestion from a confident wrong one.

I use AI every day, for reviewing code, exploring unfamiliar APIs, drafting tests and docs, and prototyping before I commit to a design. When I put AI into software, four rules keep it dependable.

  • The model sits behind an interface

    Application code talks to a small, typed abstraction, never straight to a vendor SDK. Models and providers can change without a rewrite.

  • Conversation state is explicit

    Each session keeps its own memory, isolated and thread-safe, so concurrent users never see each other's context.

  • Every call is a remote dependency

    Timeouts, retries and fallbacks. A slow or failed model response degrades one feature; it doesn't take the system down.

  • Outputs are checked, not trusted

    Structured output is validated against a schema before anything acts on it. If it doesn't conform, it doesn't go through.

Open source

Nine packages on NuGet, npm and PyPI. Download counts are live.

Open-source packages by Abhishek Joshi
PACKAGE DESCRIPTION STACK DOWNLOADS
# AI and Microsoft APIs
ResearchAI Azure OpenAI chat completions for .NET, with thread-safe per-session memory and image analysis through Azure Vision. .NET, Azure OpenAI, Azure Vision ResearchAI NuGet downloads
MSGraph.Mail.Client Reads and sends Microsoft 365 mail from .NET through the Microsoft Graph API. .NET, Microsoft Graph MSGraph.Mail.Client NuGet downloads
msgraph-email The Python counterpart: reads and sends Office 365 or Exchange mail through Microsoft Graph. Python, Microsoft Graph msgraph-email monthly PyPI downloads
IGGraphClient Fetches Instagram profiles, media and comments from .NET through the Instagram Graph API. .NET, Instagram Graph API IGGraphClient NuGet downloads
# Real-time and messaging
Kafka.SignalR Delivers Kafka messages to SignalR clients in real time. .NET, Kafka, SignalR Kafka.SignalR NuGet downloads
MQDotnetClient Wraps the official IBM MQ client for .NET, with Polly retries and transient-fault handling built in. .NET, IBM MQ MQDotnetClient NuGet downloads
# Front end
ngx-iso-form Generates Angular reactive forms from any XSD schema, such as ISO 20022 payment messages. Angular, TypeScript ngx-iso-form npm downloads
ngx-xml-message Renders XML messages as readable, read-only Angular forms. Angular, TypeScript ngx-xml-message npm downloads
sitePage.js Builds full-screen, section-by-section scrolling websites, with Bootstrap support. TypeScript, JavaScript sitePage.js npm downloads

Stack

What I reach for, from the model to the screen.

{
  "ai": ["Azure OpenAI", "Azure AI Vision", "LLM integration", "Structured outputs", "Prompt design"],
  "backEnd": ["C#", ".NET", "Python", "REST APIs", "Polly"],
  "frontEnd": ["Angular", "TypeScript", "JavaScript", "HTML and CSS"],
  "dataAndMessaging": ["Kafka", "SignalR", "IBM MQ", "Microsoft Graph", "ISO 20022"],
  "cloud": ["Azure", "Azure Static Web Apps", "GitHub Actions"],
  "yearsShipping": 16
}

Let's talk

Building something with AI, or want AI in something you've already built? I'm always up for a conversation, whether it's a role, a project or an idea you want to think through out loud.

I'm in Perth, Australia (AWST, UTC+8) and happy to meet remotely.