@dineshstack
Senior Full Stack Developer · Building SaaS products & teaching Laravel/React · 10+ years experience · Founder of Orion360 · Based in Dubai, UAE.
Make code style and static analysis automatic gates. No more "I forgot to run Pint" in review. No more type errors reaching production. Here's how to configure a Code Quality job that blocks merges — including the PHPStan baseline trick for legacy codebases.
Want to contribute to an open-source AI trading bot? Here's how — where help is most needed (strategy, ML, dashboard, DevOps), how to set up the project locally on testnet, and how to submit your first pull request. All skill levels welcome; non-code contributions count too.
A step-by-step guide to deploying an open-source AI crypto trading bot to your own VPS: prerequisites, the exact API keys you need, training the ML model, starting it with systemd, and an honest running-cost breakdown. Runs on Binance testnet in about 20 minutes.
FeaturedA deep dive into how an open-source AI crypto trading bot built on Claude works — the multi-agent LLM pipeline, the machine-learning ensemble, layered risk management, and an honest look at what the backtests actually show. Built with Python, Next.js, and the Anthropic API.
FeaturedHiggsfield's "unlimited" only works in the web UI — the CLI and API always bill credits. Here are the real per-clip costs, a plan-sizing calculation, and exactly when to use each path.
SQLite hides MySQL-specific bugs until production. Here's exactly how to add a MySQL 8.4 service container to GitHub Actions, override your test env vars, and make Pest hit a real database — including the Pusher/broadcast gotcha that will wreck your first attempt.
Step-by-step: create a .github/workflows/api-ci.yml file that installs PHP 8.4, runs Composer, and executes your Pest test suite on every push. Takes 15 minutes and works on a free GitHub account.
CI/CD sounds like DevOps buzzword soup until the day you push broken code to production at 11pm. After 10 years of making that mistake, here's what it actually means and why it will save your weekends.
Generating a voice response is the last step in the pipeline — but it's also the one that's most exposed to runaway costs. Each TTS call charges per character. Each STT call charges per second. Claude charges per token. Multiply by thousands of users and the bill can spiral fast. In Part 4, we build the Cartesia TTS service for Arabic/English voice synthesis, then add the cost tracking and rate limiting layer that keeps your API spend under control at per-user granularity.
A security-first guide to using Claude AI on a production server: guardrails, human-in-the-loop sudo, and how the AI found a real exposed API in hour one.
The hardest part of building a voice assistant isn't the speech — it's getting the AI to take the right action reliably. Claude's tool use API lets you define functions the AI can call (like get_fare_estimate or book_ride) and forces it to return structured data instead of making things up. In Part 3, we design the system prompt for Arabic/English responses, define 15+ tool schemas, and im plement the two-turn tool use flow that's required whenever Claude needs to execute an action before it can.
Arabic speech recognition has a dirty secret: most models are trained on Modern Standard Arabic (MSA) and fail on Gulf dialects like UAE or Kuwaiti Arabic. Deepgram Nova-3 is the first model to handle Gulf dialect reliably, and its multi-language mode lets users seamlessly switch between Arabic and English mid-sentence — which is exactly how people in the UAE actually speak. In Part 2, we build the SpeechToTextService that transcribes voice input, detects the language automatically, and feeds.
