Raj Kapadia

I am an engineer and music lover, passionate about AI Agents/Chatbots and overall software development.

[Raj Kapadia @RaajKapadia]

#vscode can now run #copilot #codex #claude sessions in isolated #git #worktrees, making it easier to compare patches without agents colliding in one checkout. The real multi-agent advantage isn’t generating more code; it’s producing parallel, testable, reviewable alternatives.

[Raj Kapadia @RaajKapadia]

Official docs: https://t.co/UOo4HT6zCT

[Raj Kapadia @RaajKapadia]

Meta says more than one million businesses now use its agents on WhatsApp and Messenger, while India has identified 762 government AI use cases. It’s building reliable, multilingual workflows with persistent state, validation, human review, and measurable outcomes.

[Raj Kapadia @RaajKapadia]

Help me find the best #mouse and #ketboard separate/combo. I am looking at #logitech mx series.

[Raj Kapadia @RaajKapadia]

One of most intresting #book I have read recently. #fyodor #crimeandpunishment

Must read.

[Raj Kapadia @RaajKapadia]

Node.js is issuing High-severity security fixes across the maintained 26.x, 24.x and 22.x lines, while end-of-life versions receive no official patches. Check your local runtime, CI configuration and Docker base images; not just package.json.

[Raj Kapadia @RaajKapadia]

A useful pattern for infrastructure agents: give every run its own disposable LocalStack environment, let the agent build and test freely, then return a reviewable deployment plan. Autonomy becomes much safer when mistakes happen inside infrastructure designed to disappear.

[Raj Kapadia @RaajKapadia]

CARS24 says its AI agents handle 1M+ monthly conversation minutes and recover 12% of previously lost leads. The important architecture isn’t just the model; it’s resumable workflows, explicit qualification, document handling, human approval, and measurable business outcomes.

[Raj Kapadia @RaajKapadia]

I built the same task manager AI agent 3 times—with #LangGraph, #OpenAI Agents SDK, and #Google ADK.

Same tools. Same storage. Very different developer experience.

The question isn’t “Which framework is best?” It’s “Which one fits your agent?”

🎥 https://t.co/zIoteT3frA

[Raj Kapadia @RaajKapadia]

Historic tweet.

[Sam Altman @sama]

language interfaces are going to be a big deal, i think. talk to the computer (voice or text) and get what you want, for increasingly complex definitions of "want"!

this is an early demo of what's possible (still a lot of limitations--it's very much a research release).

[OpenAI @OpenAI]

Introducing GPT-Red

An internal automated red teamer on a mission to find our models’ prompt injection vulnerabilities at scale, helping us build stronger defenses before wider deployment.

https://t.co/GxnmxxcpSk

[Raj Kapadia @RaajKapadia]

I built a personal task manager AI agent using Python and the OpenAI Agents SDK.

It can add, list, complete, and reopen tasks—and the video shows every tool call, input, and output.

Watch: https://t.co/3GLGKwpoms

#Python #AIAgents

[Raj Kapadia @RaajKapadia]

I built an AI Object Remover using OpenCV 5 + diffusion inpainting.

Mark the unwanted object, generate a mask, and let OpenCV fill the area automatically.

Video: https://t.co/s1fAmLjnNL

#OpenCV #Python #ComputerVision

[Raj Kapadia @RaajKapadia]

Not every chatbot is an AI agent.

A chatbot talks.
An AI agent does.

In this video, I explain AI agents simply - from chatbots to assistants, tools, memory, planning, workflows, and multi-agent systems.

Watch here: https://t.co/I5MGUddHOl

#AIAgents #AgenticAI #AI

[Raj Kapadia @RaajKapadia]

I built a Brand UI Generation AI Agent.

It takes brand screenshots, analyzes them, generates UI concepts, creates a new web UI, evaluates it, and improves the prompt in a loop.

Not just “generate once and hope.”
Actual AI workflow.

Watch the video here: https://t.co/Uf6oLVZWAn

[Raj Kapadia @RaajKapadia]

@yahyahishammm https://t.co/KhIsqP7AfJ

[Raj Kapadia @RaajKapadia]

RAG explained simply:

Normal AI answers from memory.

RAG first searches the right information, gives it to the AI, and then the AI answers.

Watch the video here: https://t.co/B9S355E8G0

#AI #RAG #LLM

[Raj Kapadia @RaajKapadia]

I trained a small decoder-only GPT-style LLM from scratch using TinyStories and ran it locally.

Just pure next-token prediction.

Dataset → training → testing → generation → FastAPI serving.

Small LLMs are not useless. They are focused.

https://t.co/3VUTiMe1PJ

[Raj Kapadia @RaajKapadia]

New video is live 🚀

I built a reusable YOLO object detection training pipeline with Ultralytics, controlled using a single JSON config.

Dataset → config → train → predictions → weights → metrics → artifacts.

Watch here: https://t.co/n4Ydvk7E5Z

#YOLO #ComputerVision #AI

[Raj Kapadia @RaajKapadia]

@Joi2James https://t.co/TzUfTfPIGA