Skip to content
View hirjoabaioan's full-sized avatar

Block or report hirjoabaioan

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
hirjoabaioan/README.md

Hi, I'm Ioan Hîrjoabă

Typing SVG

Website CV Email LinkedIn


🚀 About me

  • 🏗️ I design, build and operate production SaaS platforms end to end, solo — schema, backend, frontend, integrations, infra, monitoring.
  • 💼 Founder of IFindSpec / Avocățescu (practice-management SaaS), StoryFuze (content creation & multi-channel publishing), Event Creator Hub and WeddingPages.
  • 🤖 I build LLM agents that act on real business data — tool calling with RBAC and human-confirmed writes, messaging bots with autonomy levels, speech pipelines — and measure them with evals.
  • 🧰 I also run my own agent development infrastructure: router → orchestrator → sub-agents with gates and cost-aware model routing, plus an autonomous healer.
  • 🔐 Security-minded by default: prepared statements, CSRF guards, ownership checks, least-privilege keys, self-run IDOR/SQLi audits.
  • 📍 Romania · remote · 🗣️ Romanian (native), English (professional)

📈 In numbers

4 platforms live 60+ modules 62 AI tools 4 open-source agent repos Solo

🤖 Open-source agent work

Repo What it shows Results
tool-calling-agent Tool registry with 3-point RBAC, writes as human-confirmed proposals, router with circuit breaker, metering, MCP server 40 evals × 4 models: 88-98%, RBAC refusal 100%
autonomy-levels-agent WhatsApp-style agent: off / suggest / auto, customer-confirmed writes, server-side identity, dedup and loop guards Prompt injection resisted 100% on 4 models
speech-pipeline-bench Provider-agnostic STT/TTS, transcript-based video editing, RO/EN benchmark Whisper vs Scribe vs Gemini vs Google vs OpenAI: WER, cost, latency
server-update-agent Weekly patching agent for my production server: policy decides (7-day soak, follow-up patch, human approval for minors), backup → test → prod with auto-rollback; the LLM only summarises changelogs Rollback drill 5/5 restored in ~1 s each, 0 HTTP 5xx

🧩 What I'm building

Project What it is Stack
IFindSpec / Avocățescu CRM, scheduling, payments, AI assistants, video & social marketing for professionals — dual-branded from one codebase PHP 8.3 · MariaDB · Node · Stripe · AI
StoryFuze Video editing by transcript, AI creatives, publishing to Meta / TikTok / YouTube / LinkedIn PHP · Node · ffmpeg · fal.ai
Event Creator Hub Paperless judging, real-time scoreboards, WhatsApp PHP · MySQL · Stripe
WeddingPages Wedding-site builder + venue partner portal PHP · Stripe Connect · S3
World Pulse ⭐ open source Real-time global events map on a zero-dependency D3 globe Node · D3 · GDELT
AI Orchestration Router, orchestrator & self-healing agent I build with (healer source published) Node · Claude · DeepSeek

➡️ Full showcase with architecture notes and screenshots: hirjoabaioan/portfolio

🛠️ Tech stack

Backend
PHP Node.js Python MariaDB MySQL REST

Frontend
JavaScript HTML5 CSS3 Bootstrap jQuery D3 PWA

Infrastructure & DevOps
Linux nginx Docker Hetzner Cloudflare PM2 Grafana Prometheus Git

Integrations & AI
Stripe Meta TikTok YouTube WhatsApp S3 FFmpeg Claude Gemini

📊 GitHub stats

GitHub stats GitHub streak

Most of my work lives in private, commercial repositories — the numbers above include private contributions.

⚙️ Engineering principles

  • Pragmatic over fashionable — framework-free PHP where it's fastest to ship, Node microservices where they fit (media, PDF, queues).
  • Own the whole lifecycle — test env → lint → functional checks → production, with backups and rollbacks; public repos run CI with offline eval replay on every push.
  • Observable by default — metrics, logs and alerts (Prometheus, Loki, Grafana → Telegram) on everything I run.

Open to full-time remote roles (EU/CET) and B2B contracts in AI agent engineering · [email protected]

Pinned Loading

  1. autonomy-levels-agent autonomy-levels-agent Public

    Messaging agent with off/suggest/auto autonomy: customer-confirmed writes, server-side identity, dedup/loop guards, human handoff, 25 eval scripts

    TypeScript

  2. portfolio portfolio Public

    Showcase of production SaaS platforms I built and operate solo.

    JavaScript

  3. speech-pipeline-bench speech-pipeline-bench Public

    Provider-agnostic STT/TTS layer, transcript-based video editor, and an honest Romanian/English benchmark (Whisper, Scribe, Gemini, Google, OpenAI)

    TypeScript

  4. tool-calling-agent tool-calling-agent Public

    Production-shaped tool-calling agent in TypeScript: 3-point RBAC, proposal-based writes, circuit-breaker router, metering, MCP server, 40-case deterministic evals

    TypeScript

  5. world-pulse world-pulse Public

    Live global map of conflicts, news, official announcements and disasters from open sources — zero-dependency Node + D3

    JavaScript