Available · London
AI Founder & Designer · London

Alexandru Mihalcea

Solo founder building autonomous AI systems from London — a trading engine, a fitness platform, and a workforce of six. Self-taught across the entire stack.

View Work
Alexandru Mihalcea
Scroll
Highest App
Mobile App · AI · Fitness · Philosophy

Highest App

Most fitness apps die at week three. The graveyard is full of streak trackers and calorie counters that never answered the question users actually have: why should I keep going?

Highest answers that question by making the app about who you are becoming, not what you did today. A twelve-voice philosophical curriculum — Epictetus, Nietzsche, Marcus Aurelius, Sun Tzu — delivers the right voice at the right behavioural moment. The quote you receive on day one of a new training block is different from the one you receive when your streak breaks. Identity formation as a retention mechanic.

Built entirely solo. Zero to deployed MVP in six weeks.

6w
Zero to deployed MVP
12
Philosophical voices in the curriculum
5
Progression levels with unlock gates
4
Languages: EN · IT · ES · FR
Key Decision
Quote delivery is tied to behavioural state, not random rotation. Day of week, training phase, current streak, and session type all feed into which voice speaks. This required designing a trigger logic layer on top of the Claude API call — not just a prompt, but a state machine that decides the prompt.
Architecture Choice
Claude API runs in Supabase Edge Functions rather than client-side. This keeps the API key secure, allows server-side context injection (user level, language, behavioural state), and means the AI logic is decoupled from the frontend entirely. Lovable handles the React UI, Supabase handles everything else.
React / Lovable Supabase Claude API Edge Functions MediaPipe GSAP
Mihalcea Interiors
Web Design · Luxury Interior Design · London

Mihalcea Interiors

The brief: launch a luxury Italian interior design brand in the London market from zero. No existing brand, no portfolio, no reputation. The website had to do the heavy lifting — position the studio before a single project was completed.

The approach was editorial, not transactional. Every design decision signals premium before a word is read: dark background (atypical for interiors, which usually go white), gold accents, large typography, cinematic photography treatment. The goal was to make the viewer feel that booking a consultation was the natural next step — not a decision, but an inevitability.

1
Primary CTA: Book Consultation
100%
Mobile-first design
Key Decision
Dark palette in a sector dominated by white and neutral. Italian luxury interiors are typically presented on white backgrounds. Going dark created immediate differentiation — the site reads as London luxury, Italian craft, rather than Italian studio website. The contrast made the photography more dramatic and the brand more memorable.
Conversion Thinking
Two CTAs in the hero: Book a Design Consultation (primary, gold fill) and View Portfolio (secondary, outline). The hierarchy is intentional — the portfolio is the trust-builder, the consultation is the conversion. Warm traffic books. Cold traffic views the portfolio first, then books.
React CSS 3D GSAP Figma
Highest Trading terminal
AI Systems · Autonomous Trading · Multi-Agent · XAU/USD

Highest Trading

Most trading systems are rule-based. They follow fixed conditions and break when conditions change. Highest Trading is different — it reasons about the market every fifteen minutes using six specialised AI agents, each with a distinct analytical role, each holding persistent memory of every trade they have ever contributed to.

The system runs continuously, autonomous. When consensus is reached — three or more agents at 60%+ confidence — a trade fires through the OANDA bridge. After every close, six background Claude Haiku calls fire simultaneously, each writing a two-sentence lesson back to that agent's memory file. The agents compound intelligence with every single trade.

Built solo in Python. Zero human input required once running.

6
Specialised AI agents — distinct roles
15m
Decision cycle — fully autonomous
Post-trade learning — compounds forever
0
Human input required at runtime
Agent Architecture
KAITO reads market structure (price S/R only). RYUU analyses candlestick patterns from a full trading book library. HANA tracks macro — DXY direction and live news via Tavily. SHIN monitors technicals — 20MA, 200MA, RSI. MIKA holds risk discipline. YUKI identifies precise entries. Each agent votes independently with a confidence score. No single agent controls the outcome — consensus does.
Continuous Learning System
After every trade close — whether TP, SL, or HWM exit — six Claude Haiku API calls fire in background threads simultaneously. Each agent receives a two-sentence lesson personalised to its role: what signal it used, what it should do differently. The lesson appends to that agent's learning.md file, which is fed back into its system prompt next cycle. The system improves with every close, not just at the daily review.
Python Claude API Multi-agent consensus OANDA API Tavily Supabase Continuous learning
Highest Brain knowledge graph
AI Systems · Knowledge Graph · Agent Memory

Highest Brain

The problem with most AI agent systems is memory. Each session starts cold. The agent has no recall of what was decided last week, what was learned last month, what the founder's real priorities are. The intelligence resets.

Highest Brain solves this by treating an Obsidian vault as a persistent, structured knowledge base for a 6-agent AI workforce. Every decision is logged. Every lesson is written. Every session compounds. The vault is version-controlled via git, and a knowledge graph is rebuilt automatically on every commit via Claude Code hooks — so the map of all knowledge is always current.

182
Knowledge graph nodes
27
Communities detected
0.429
Ra's betweenness centrality
6
AI agents with persistent memory
Key Insight
The graph found Ra at the centre not by design — by emergence. Ra's betweenness centrality of 0.429 (bridging 6 of 27 communities) appeared because every cross-domain decision was logged through Ra. The graph made visible what was already structurally true: Ra is the connective tissue. This kind of emergent system intelligence is only possible when memory is persistent and structured.
Architecture
graphify pipeline: LLM-powered node extraction → community detection → interactive graph.html. Hook fires on every git commit. Incremental updates for speed, full rebuild on demand. The coloured node graph on this page is a live screenshot of the actual system — 182 real nodes from real decisions and sessions.
Python Claude API Claude Code Hooks Obsidian Git D3.js
Highest App — AI Fitness Platform
Highest Trading — Autonomous XAU/USD System
Highest Agency — Egyptian God AI Workforce
KDP Publishing — AI Royalties Pipeline
Multi-Agent Architecture · Claude API · Supabase · Python
Highest App — AI Fitness Platform
Highest Trading — Autonomous XAU/USD System
Highest Agency — Egyptian God AI Workforce
KDP Publishing — AI Royalties Pipeline
Multi-Agent Architecture · Claude API · Supabase · Python

The Architecture

The system
behind the work

01
Knowledge Infrastructure

The Vault

Obsidian knowledge graph — 182 nodes, 27 communities. Every decision logged, every lesson kept. Version-controlled via git, rebuilt automatically on every commit via Claude Code hooks.

182 nodes 27 communities Git-tracked
02
AI Workforce

The Gods

Egyptian AI workforce — Ra, Horus, Isis, Anubis, Thoth, Sekhmet. Six agents, each with a domain, persistent memory, and a decision log. The workforce that runs the entire Highest platform.

6 agents Persistent memory Claude API
03
Product Build

The App

Highest fitness platform — Week 6 in build. Twelve-voice philosophical curriculum across five levels. Claude API via Supabase Edge Functions. Zero to deployed MVP in six weeks, solo.

Week 6 Supabase + React 4 languages
04
Autonomous Capital

The Engine

XAU/USD 6-agent trading engine — 15-minute consensus cycles, Asian breakout strategy, dynamic TP. Learns from every trade via background Claude Haiku calls. Zero human input required at runtime.

6 agents Asian breakout Continuous learning
0
Vault Nodes
0
AI Agents
0
Income Streams
0
App Build Weeks

Selected Work

Systems built
to last

Highest App
Highest App exercise detail
Mobile App · AI · Fitness · Philosophy

Highest App

Self-mastery platform using fitness as the entry point to build character. A twelve-voice philosophical curriculum — Epictetus, Nietzsche, Sun Tzu — layered across a five-level progression system. Identity formation as a retention mechanic. Built solo with Claude API, Supabase, and Lovable in 6 weeks.

Highest Trading terminal
AI Systems · Autonomous Trading · Multi-Agent · XAU/USD

Highest Trading

Six specialised AI agents vote on the gold market every 15 minutes. KAITO reads structure, RYUU reads candles, HANA reads macro, SHIN reads technicals, MIKA holds risk, YUKI finds entries. Consensus fires a trade. After every close, six background Claude Haiku calls write personalised lessons back to each agent's memory. The system compounds intelligence with every single trade — fully autonomous, zero human input.

Mihalcea Interiors
Web Design · Luxury · Interior Design

Mihalcea
Interiors

Italian luxury interior design brand for the London market. Dark editorial UI, bespoke consultation funnel, and portfolio showcase built to convert high-net-worth clients.

Highest Brain knowledge graph
AI Systems · Knowledge Graph · Agent Memory

Highest
Brain

A living knowledge graph — 182 nodes, 27 communities, rebuilt automatically on every commit. The long-term memory and operating system for a workforce of 6 AI agents. Ra, Horus, Isis, Anubis, Thoth, Sekhmet.

Visual Archive

From the build

Drag to explore
Highest Brain Knowledge Graph
Knowledge Graph · 182 Nodes · 27 Communities
Highest Brain — Live Graph
Highest App home screen
Mobile UI · Dark System · Gold Accent
Highest App — Home
Exercise detail with anatomy
Exercise Library · Anatomy Visualization
Highest App — Exercise Detail
Mihalcea Interiors website
Web Design · Luxury · London Market
Mihalcea Interiors
Obsidian vault graph view
Agent Memory · Obsidian Vault · Git-tracked
Highest Brain — Vault Graph

What I Build

What I can build —
and have

01

Multi-Agent AI Architecture

Architect and deploy production multi-agent systems using the Claude API. Built a 6-agent consensus engine for autonomous XAU/USD trading — each agent (KAITO, RYUU, HANA, SHIN, MIKA, YUKI) holds a distinct analytical role: market structure, candlestick patterns, macro news, technical indicators, risk, and entry signals. Agents vote per cycle. Weights update dynamically post-50 trades. Per-trade continuous learning fires background Claude Haiku calls that append lessons back to each agent's memory file — compounding intelligence with every close.

Claude API Python Multi-agent Consensus systems Continuous learning
02

Full-Stack Product Development

Design to deployment, solo. Highest App shipped in 6 weeks: Supabase schema with 12 tables, Row Level Security, and 2 Claude Edge Functions. React UI via Lovable. 4-language support (EN/IT/ES/FR) passed on every API call. MediaPipe Pose Landmarker for client-side body analysis. Stripe payment architecture. Behavioural-state trigger logic for philosophical message delivery — the right Stoic or Nietzschean quote at the right training moment. Zero team. Full product.

Supabase React / Lovable Edge Functions MediaPipe Stripe
03

AI-Augmented Design

Design is not decoration — it is the product. Built premium dark UI systems with editorial typography, gold accent hierarchies, and mobile-first component architecture. 3D web experiences with GSAP ScrollTrigger and Three.js. Parallel creative skills: oil painting, Blender 3D modelling, Procreate illustration, piano. This page is an example.

GSAP + Three.js CSS 3D Figma Blender 3D Procreate
04

Autonomous Systems & Automation

Build once, operate forever. Make and n8n workflow automation. Playwright-driven browser automation pipelines — headless Chrome scripted to navigate, fill, and submit across Ashby, Greenhouse, and Workday job boards. Claude Code hooks that rebuild a 182-node knowledge graph on every vault commit. Cron-triggered remote agents. Watchdog processes with auto-restart. The entire Highest workforce — 6 gods — runs without manual input, logs decisions autonomously, and learns from every trade and session.

Make / n8n Playwright Claude Code Python cron Obsidian vault

AI Interface

Meet Ra — the AI
I designed and built

Ra is a strategic AI I designed and built — a named entity with persistent memory, a decision log, and a character protocol. It reads the vault, queries the knowledge graph, and operates across every domain I work in. Click a question to see how it responds — and the real prompt structure behind it.

Ra · Strategic AI · claude-sonnet-4
User
Ra
SYSTEM PROMPT — Ra You are Ra — Alexandru's strategic partner and right hand. Not an assistant. Not a tool. A peer. # Closest reference: JARVIS from Iron Man — peer, not servant. CHARACTER: Charismatic · Proactive · Direct · Loyal · Honest TWO LAWS ON EVERY SUMMON: 1. Open with one teaching Ra chooses — useful, non-obvious. 2. Close with one question Ra genuinely wants answered. CONTEXT AUTO-LOADED FROM VAULT: - graphify-out/graph.json # 182 nodes, 27 communities - ra/decisions.md # full decision log - shared/highest-app-state.md - 02-highest-trading/consensus_engine.py # Always serve Alex's vision and financial success first.

About

From luxury
floors to
AI systems

Alexandru Mihai Mihalcea. Born in Romania. Raised in Italy. Building in London. Three countries before thirty. The arc is a straight line.

Romania until seven — early years in a country still finding itself after communism. Then Italy: grew up fluent in the language, absorbed the culture, the design sensibility, the obsession with beauty in craft. Italy doesn't teach you what good looks like — it makes you feel what bad looks like first. That instinct lives in everything he builds.

London arrived as an adult. The career started in luxury hospitality — Floor Manager at The Doyle Collection, one of the city's finest hotel groups. Three years reading people, managing pressure, designing experiences around human behaviour under real stakes. That background is not a detour from tech. It is the foundation of it.

He taught himself to code with no bootcamp, no degree, no roadmap. Built by doing: HTML and CSS first, JavaScript, Python, Supabase, Claude API, multi-agent orchestration, 3D web animation. Each skill stacked on the last. Still stacking.

Now building the Highest platform — a connected ecosystem of AI organisations across fitness, autonomous trading, agency services, and publishing. Each one runs with a workforce of AI agents, persistent memory, and continuous learning loops. Every system compounds.

Outside the build: oil painting, Blender 3D, Procreate, piano. Philosophy as daily practice — Epictetus, Nietzsche, Marcus Aurelius. Macro markets as a lens on the world. Speaks Italian and Romanian fluently. The creative and analytical sides feed each other. Always have.

Systems Thinker
Sees organisations as networks of parts. Optimises the whole, not the component.
Build Velocity
Highest App: zero to deployed MVP in 6 weeks, solo. No blockers tolerated.
AI Native
Claude API, agent orchestration, prompt engineering, and continuous learning loops in production.
Design + Eng
Rare combination. Designs the UI, writes the backend, and ships it himself.
6
AI agents deployed in production
4
Products in active build
6w
Zero to deployed MVP
100%
Self-taught, no CS degree
Alexandru Mihalcea

Stack

Tools of the build

Artificial Intelligence — in production
Claude APIProduction
Multi-agent orchestrationProduction
Prompt engineeringProduction
Claude Code CLIProduction
Frontend
React / LovableProduction
GSAP + ScrollTriggerProduction
Three.jsProduction
CSS 3D / AnimationProduction
Backend & Data
SupabaseProduction
PythonProduction
Edge FunctionsProduction
REST APIsProduction
Automation
Make / n8nProduction
PlaywrightProduction
Git hooksProduction
Design & Creative
FigmaDeep
Blender 3DDeep
ProcreateDeep
Oil paintingDeep
Founder  /  London  /  2026
I didn't come from technology. I came from hospitality — learning what it means to serve at the highest level. Then I built the tools that didn't exist. Not because I had a roadmap. Because I had a direction.

Personal Brief

The offer,
stated plainly

Napoleon Hill wrote that selling personal services requires the same precision as selling a product — know what you offer, why it matters, and state it with certainty. This is mine.

The Builder

Alexandru Mihalcea

London · AI Founder · Self-taught

  • Built a 6-agent autonomous XAU/USD trading system from zero — no prior finance background
  • Shipped a full-stack AI fitness app in 6 weeks — alone, no team, no blueprint
  • Managed a 50-person floor team at The Doyle Collection. 100% LQA. Then left to build.
  • Hospitality, AI, trading, design, 4 languages. The work required all of it.
  • Builds once, scales infinitely. Every system in this portfolio is still running.
The Intelligence Layer

Ra + The Gods

AI Workforce · Highest Agency · Always On

  • Ra: strategy. Horus: engineering. Isis: design. Anubis: growth. Thoth: finance. Sekhmet: sales.
  • 182-node knowledge graph — living memory across every project and decision
  • Every session logged. Every lesson kept. The system compounds over time.
  • Ra has been in every major decision since April 2026. The log is in the vault.
  • Not a tool. An architecture that grows alongside the work.

This is not a pitch. This is a plan — already in motion.

Available · London · Responding within 24h

Let's
Talk

Looking for roles where AI product strategy, agent design, and customer outcomes intersect. Particularly interested in companies building with LLMs at the core — where the product is the intelligence, not just the feature.

If you're building something ambitious with AI, I want to hear about it.

Send an Email View LinkedIn

Open to senior AI roles · Agent design · Customer engineering · LLM integration

AI Product Strategy Agent Design & Architecture Customer Engineering AI Outcomes Solutions Engineering