Hello, I'm
Building practical tools at the intersection of Python, AI, and automation.
Currently expanding into full-stack development.
I'm a self-taught developer based in the Philippines. I started programming to solve real problems — building AI-powered tools, automating repetitive workflows, and experimenting with local LLM setups on my own GPU rig.
I work mostly in Python and am actively expanding into full-stack development. I enjoy building things that are actually useful: automation pipelines, utility libraries, and multi-agent AI systems. When I'm not coding, I'm tinkering with new tools or running experiments on my local machine.
name = "Rolly Calma"
role = "Python Developer"
focus = "AI & Automation"
location = "Philippines 🇵🇭"
building = "AI analysis bot 🔨"
status = "open_to_work"
Scripts, pipelines, and bots that run unattended — from file processing workflows to multi-step API integrations.
Multi-agent systems, local LLM setups, and practical AI tools using Ollama and Claude API — production-ready, not just demos.
Flask backends, REST APIs, and real-time streaming apps with SSE — from working prototype to deployed and running.
14 zero-dependency Python modules, each drop-in ready. Covers thread-safe rate limiting, retry with full-jitter backoff, structured logging, async helpers, and HMAC signing — tested in workflows handling hundreds of API calls per session. Zero external dependencies, pure stdlib only.
Runs 2–6 AI agents simultaneously in one live conversation. @mention any model to pull it in — mix local Ollama (sub-8B, runs on modest hardware) with Claude API in the same thread. Built-in debate mode auto-runs 3-round structured discussions. SSE streaming delivers responses token-by-token with no page refresh.
Fully automated end-to-end pipeline — processes and schedules up to 4 uploads/day via YouTube Data API v3. Handles watermarking, loudness normalization, and optional BGM mixing via FFmpeg, then moves files to done/ automatically. Saves ~2–3 hours of manual work per week. Runs on Windows startup via Task Scheduler.
Two battle-tested utilities pulled from production code. A thread-safe token-bucket RateLimiter supporting configurable call budgets per second, plus a @retry decorator with full-jitter exponential backoff — handles hundreds of API requests per session reliably. Zero dependencies, drop-in for any project.
20 merged PRs across 9 repos — bug fixes, new features, tests, and refactors.
I'm always open to interesting projects, collaborations, or just talking tech.