BUILD_VERSION: 13.0.AGENTS

37.2513643° N, -115.8452064° W

Răzvan Frandeș.

12 years in web development. The stack changed,
the thinking stayed.

Lately I build with teams of AI agents, and the systems that keep them honest.

CORE_STACK

Expertise centered on high-performance runtimes and scalable data architectures.

  1. Runtimes & Logic

    • TypeScript
    • Next.js
    • Node.js
    • Bun.js
    • Hono.js
    • Remix.js
    • TanStack
    • Rust
    • Python
  2. State & Data

    • React Query
    • Zustand
    • SWR
    • GraphQL
    • REST API
    • WebSockets
  3. AI & Agents NEW

    • Claude Code (skills · subagents · workflows)
    • MCP servers
    • Anthropic API (tool use · caching)
    • OpenAI API
    • LLM-as-judge evals
    • Speech AI (Whisper · XTTS)
  4. CMS

    • Strapi
    • WordPress Headless
    • WordPress (ACF, custom themes)
  5. Database

    • PostgreSQL
    • Supabase
    • MySQL
    • Redis
    • SQLite
  6. Infra & Delivery NEW

    • Cloudflare Workers & Workflows
    • GitHub Actions
    • Docker
    • Playwright
    • Vitest
  7. Styling & Animation

    • Tailwind CSS
    • GSAP
    • three.js
    • GLSL shaders
  8. Design & Prototyping

    • Figma
    • Adobe Illustrator
    • Adobe Photoshop
[ STATUS: AGENTS_ONLINE ]

Agent_Systems

How I build now. I design the system, a team of agents does the work, and other agents try to break it.

Skip the diagram
Diagram of an agent workflow: an orchestrator fans a task out to five builder agents working in parallel. Critics reject three of the outputs. Three verifiers check what is left and two of the three must agree. One survivor merges back into the result.
  1. IN_USE

    # SYSTEM_ID: AGENT_WORKFLOWS

    Multi-Agent Workflows

    I run multi-agent workflows in Claude Code: parallel builders, judges, critics and adversarial verifiers. A claim survives only if independent agents can't break it. Builders take their own headless-Chrome screenshots, look at them and iterate.

    > OUTPUT: 72 workflow scripts so far. A 109-agent line-by-line audit confirmed 51 bugs, all fixed test-first. A 76-agent audit checked a trading engine against the exchange docs, with 2 of 3 verifiers required to agree.

    • Claude Code
    • Subagents
    • Workflows
    • Puppeteer
    • Adversarial review
  2. v0.9_BUILD

    # SYSTEM_ID: MT_KIT

    MT.Kit

    A context warehouse for Claude Code: 56 skills that take a product from a raw signal to PRD, architecture, test-first implementation and review. Task state lives on a GitHub Projects board and changes only through mtk, a TypeScript CLI built for agents to call.

    > OUTPUT: Tested end to end by agents against an emulated GitHub, two simulated users and a simulated human. Seven test phases, every finding fixed and confirmed.

    • Claude Code skills
    • TypeScript CLI
    • GitHub Projects v2
    • Vitest
    • Evals
  3. M6_DONE

    # SYSTEM_ID: STRUDEL_RUNNER

    Strudel Runner

    Live-coded music where Claude Code is the composer. A custom MCP server with 13 tools plays, edits and exports Strudel patterns. Edits hot-reload into a browser engine with MIDI out to hardware and soft synths.

    > OUTPUT: A council of five specialist agents argues the arrangement until every one of them signs off.

    • MCP SDK
    • TypeScript
    • Strudel
    • Web Audio
    • WebMIDI
  4. LOCAL

    # SYSTEM_ID: COMPOSE_DSL

    Compose: a score language for an AI composer

    Four bars of MIDI as raw JSON is about 1,600 lines, too fragile for an agent. So I designed a compact .music score language, a compiler into the sequencer's format and a Claude Code skill. I talk, Claude writes the score, the sequencer plays it.

    > OUTPUT: Round-trips through the app's own deserializer and MIDI export.

    • DSL
    • Compiler
    • React
    • MobX
    • Claude Code skill
  5. PROTOTYPE

    # SYSTEM_ID: AGENT_TEAMS

    Agent Teams from Bash

    Headless claude -p agents chained from bash, each given only the context it needs. Team Factory scouts SaaS ideas from a team's real skills and runs each through ten validator agents, built on Stafford Beer's Viable System Model. Roundtable puts 16 expert personas in one transcript to debate a topic, with me as moderator.

    > OUTPUT: Structured JSON state per agent, every validation step committed to git.

    • Claude CLI
    • Bash
    • Context engineering
    • JSON state
  6. IN_USE

    # SYSTEM_ID: GUARDRAILS

    Guardrails as Skills

    Agents love a good backtest. I wrote a skill that won't let them report one until data validation, bias guards and multiple-testing corrections pass, each rule taken from a real failure. Plus a governed research workspace: hypotheses pre-registered, every trial counted.

    > OUTPUT: 1,096 curated agent skills behind activation profiles and a static security scanner. 4 quarantined.

    • Claude Code skills
    • Python
    • SPA/StepM
    • CPCV
    • Supply-chain scanning
[ STATUS: COMPILED_SUCCESSFULLY ]

Case_Logs

# SYSTEM_ID: NEURAL_MONITORING_ECOSYSTEM

v1.0_PROD

Neural Monitoring Ecosystem

Built a neural monitoring app for clinical psychologists. The Neurosity headset throws 20MB of data per minute. Implemented spike detection algorithms from medical research. After a session, the algorithm extracts relevant moments and sends only those to AI for interpretation — not the raw flood, just what matters.

> OUTPUT: Delivered and in use. Reduced interpretation time from manual review to automated AI analysis.

  • Headless CMS
  • REST API
  • TanStack Query
  • React.js
  • React Router
  • Digital Signal Processing
  • Neurosity SDK
OPEN CASE FILE

# SYSTEM_ID: PREDICTION-MARKET_TRADING_STACK

RESEARCH_CLOSED

Prediction-Market Trading Stack

Real-time trading on Polymarket's 5-minute crypto markets. A Bun platform fuses price feeds from four exchanges with order-book streams, signs EIP-712 orders and redeems on-chain on Polygon. A Rust engine took over the hot path: 16 crates, signing in about 1 ms, order hashes byte-identical to the JS signer. Around it: on-chain wallet forensics, a copy-trading desk, and a research lab on a 1.7 GB lookahead-free cache with a Rust backtester.

> OUTPUT: Agent swarms tested about 1,500 hypotheses under pre-registered rules. The honest result: the market is efficient after fees. Bots paused. The rigor stayed.

  • Bun
  • TypeScript
  • Rust
  • tokio
  • Polygon
  • WebSockets
  • SQLite
  • Polars
  • Claude Code workflows
OPEN CASE FILE

# SYSTEM_ID: AI_TUTOR

v1.0_PROD

AI Tutor

Built with a friend three years ago. I handled the frontend (React, TypeScript, Zustand), he did the backend in Python, database on Supabase. AI tutor for Romanian students, grades 1-8. Each student picks a subject and gets guided through lessons, exercises, and explanations. The challenge: GPT-4 back then couldn't do basic math reliably. We had to build our own logic layer to handle calculations and validate answers before showing them to students. Learned a lot about where AI breaks and how to build around it.

> OUTPUT: Working product, still running. First real experience with AI's edges.

  • Python
  • React.js
  • React Router
  • TypeScript
  • Supabase
  • Strapi
  • Zustand
  • Tailwind CSS
  • Headless UI
OPEN CASE FILE

# SYSTEM_ID: PIMP_MY_SAMPLE

v0.4.0_LIVE

Pimp My Sample

Chrome extension for music producers: record any browser tab into a real sample library on your disk. Capture runs in an offscreen AudioWorklet, so recording survives closing the panel. Key, BPM and loudness analysis runs in a worker, with a hand-written Goertzel chromagram for key detection. Trim, chop on onsets, export lossless WAV.

> OUTPUT: Published on the Chrome Web Store. Built in parallel agent waves in isolated git worktrees, test-first for the logic.

  • Chrome MV3
  • WXT
  • React 19
  • TypeScript
  • AudioWorklet
  • File System Access API
  • Vitest
  • Playwright
OPEN CASE FILE

# SYSTEM_ID: DUBAI_HOSPITALITY

Dubai Hospitality

Three new hotel websites built from scratch. Pixel-perfect execution, complex animations: curtain reveals, parallax scrolling, staggered menu transitions, grid filtering with row animations.

> OUTPUT: Live websites for luxury hotel brands. Complex animation systems. Cleaner, faster codebase.

  • Next.js
  • TypeScript
  • Tailwind
  • Optimizely CMS
  • GSAP
  • A/B Testing
OPEN CASE FILE

# SYSTEM_ID: CODE_OF_MAAT

FOUNDER

Code of Maat

My agency: software, AI and design. Websites, web and mobile apps, AI integrations. I built our starter theme (Vite, Tailwind v4, an ACF page builder) and the delivery pipeline: GitHub Actions with lint, type-check, build, atomic deploys with rollback and route checks, behind Cloudflare. Eight client sites built on it so far: clinics, a music school, artists and cultural organisations.

> OUTPUT: Multilingual sites, schema.org graphs, llms.txt and WebGL heroes with fallbacks, plus a GDPR-grade patient portal with magic-link login and private uploads. The agency's own site is designed with multi-agent design pipelines.

  • WordPress
  • ACF
  • Tailwind v4
  • Vite
  • three.js
  • GSAP
  • GitHub Actions
  • Cloudflare
  • Playwright
OPEN CASE FILE

# SYSTEM_ID: B2B_COMMERCE_ENGINE

B2B Commerce Engine

Large B2B platform. Multiple user roles: suppliers, buyers, admins. Product management with certificates, MOQ logic, tiered pricing. Polling-based chat, order tracking, approval workflows, email notifications, multilanguage support.

> OUTPUT: Full system built from scratch. Complex architecture, multiple user roles, real-world business logic.

  • Next.js
  • Headless CMS
  • Zustand
  • TypeScript
  • Tailwind CSS
OPEN CASE FILE
[ 5 SIGNALS RECEIVED ]

Testimonials

  1. Rune — CEO, Include ApS

    Razvan is the kind of developer you want on your team: technically sharp, genuinely positive, and he gets things done. He's quick to pick up new tools and thinks about AI and modern tech in a way most developers don't yet.

    Responsive, collaborative, and reliable. Would work with him again without hesitation.

  2. Andrei Puni — CEO, Voxcode

    I had the pleasure of working with Răzvan, and I can't recommend him highly enough. As a frontend developer, he stands out for his meticulous attention to detail and pixel-perfect work.

    Răzvan is not just about getting the job done; he truly understands the business needs and implements effective solutions with a proactive approach. His expertise in React and Next.js is evident in the high-quality, efficient, and well-organized code he delivers.

    One of his remarkable qualities is his dedication to reducing unused JavaScript and CSS, optimizing assets, and ensuring fast, resource-efficient products. You can see that he is truly passionate about web development.

    Besides his technical skills, Răzvan is also fun to work with, making the collaboration enjoyable and productive. If you're looking for someone who combines technical excellence with a great work ethic, Răzvan is the developer you need.

  3. Cosmin Tutunaru — AlgotechSolution

    Working with Răzvan was smooth and efficient. He understood what we needed quickly, proposed clear solutions, and delivered a clean, professional result. Highly recommended.

  4. Dacian — Dual Shaman

    I needed a Chrome extension with native audio processing, not exactly a standard project. Răzvan understood right away what I wanted, came up with elegant solutions and delivered something that went beyond what I expected. The audio visualisations are fluid, the extension is stable, the code is clean. Exactly the kind of developer who makes interesting things look easy.

  5. Adrian Frandes — Psychologist

    The website Răzvan built represents exactly what I want to convey as a psychologist. Designed from scratch, thought through carefully, no generic templates. You can tell he understood my personal brand instead of just starting to build. Very pleasant to work with.

R&D Labs

Exploratory code & algorithmic research

Map of the eight labs. Lines join labs that share a technology.

  1. AI_Psy

    MindMapper AI

    Multi-agent system for psychological pattern analysis. Doesn't give advice — helps users understand their own mind.

    • Claude API
    • React
    • Multi-Agent
  2. Chrome_Ext

    JSON Formatter+

    Chrome extension upgrade. Added Copy Object, Copy Minified, Copy Key, Copy as TypeScript, Copy Path, and tree collapse controls.

    • Chrome Extension
    • JavaScript
  3. Privacy

    ChatGPT Data Extractor

    Chrome extension that exports your ChatGPT conversations locally. Built my own because existing extensions sell user data to third parties.

    • Chrome Extension
    • Local Storage
  4. Arch_Exp

    Monday.com as DB

    Experimental headless architecture using Monday.com as a GraphQL data source for a Next.js frontend.

    • Next.js
    • GraphQL
    • Monday API
  5. Live_App

    HOPE Wellness

    PWA for wellbeing exercises with progress tracking, journaling, and push notifications. WordPress headless + React. 500+ users.

    • React
    • PWA
    • WordPress Headless
  6. AI_Research

    YieldMap

    Real-world-asset yields compared for retail investors. A hand-written Claude tool-use agent researches each product with five tested tools, and its output waits in a queue for human approval. Prompt caching cut its input cost by about 80%.

    • React Router 7
    • PostgreSQL
    • Drizzle
    • Anthropic SDK
    • Bun workers
  7. Audio_Viz

    VJ Visual Engine

    Browser VJ engine for live techno. Clips become abstract, audio-reactive visuals through custom shaders, 1,754 MilkDrop presets and Hydra patches. Every parameter can follow the kick, bass, mids or hats. Records to WebM.

    • three.js
    • GLSL
    • Butterchurn
    • Hydra
    • Web Audio
  8. Agent_Loops

    Autonomous Builds

    Long-running coding agents on real specs: a Go headless CMS built task by task by a Ralph loop from a PRD, and a Bun CMS split into 124 features on an agent board.

    • Go
    • Bun
    • PostgreSQL
    • Ralph loop
    • Agent boards

THE_THESIS

Code is transient. Architecture is permanent.

In an era of LLM-generated syntax, the engineer's value shifts from writing loops to orchestrating resilient systems.