Workflow design
Find the moments where AI can remove friction, then shape a workflow people will actually use.
01 / systems in motion
Design the AI layer between intent and action.
Practical AI consulting for teams that want work to feel clearer, calmer, and more capable.
human intent → useful system
Scroll to move through the system
02 / the principle
Make the invisible legible.
Good AI feels less like a feature and more like a new rhythm: the right context arrives, the next decision gets lighter, and the system stays understandable.
03 / services
From a sharp question to a system people trust.
Custom applications
Build focused copilots, dashboards, and automations around the data and decisions a team owns.
Prototype to production
Turn an impressive demo into a reliable, observable system with a sensible operating model.
04 / project ledger
Systems built to make work move.
A privacy-safe public snapshot of selected work. Sensitive product details and private project names stay out of the portfolio.
Active buildSYS / 01
Agent OS / Olympus
Local-first AI workspace for teams, agents, tools, memory, and production artifacts.
Stack: Next.js, TypeScript, Node/Express, Socket.IO, Prisma, SQLite, React Flow. Current: active local development with evidence and runtime gates still in progress.
Staging candidateSYS / 02
Iisi platform
Privacy-first workspace concept for document-heavy financial operations and review.
Stack: Next.js, TypeScript, Supabase/Postgres, Row Level Security, Playwright. Current: synthetic-data staging candidate; not approved for real customer data.
Runtime / prototypeSYS / 03
Auto Company
Autonomous AI-company loop with role-based agents for research, coding, deployment, and marketing.
Stack: Codex CLI, Claude Code, Bash, Make, Windows/WSL. Current: local runtime and automation patterns; no public deployment is claimed.
Active companionSYS / 04
Hermes Workspace
Agent command center for chat, files, memory, skills, MCP, terminal, operations, and swarm workflows.
Stack: React, TypeScript, Vite, TanStack Router, Electron, Framer Motion, Three.js. Current: mature companion codebase with ongoing feature work.
Local utilitySYS / 05
Model routing utility
Local proxy layer for model discovery, request translation, and usage-aware routing.
Stack: Node.js, JavaScript, Express, HTTP and OpenAI-compatible adapters. Current: local utility; not advertised as a hosted service.
LiveSYS / 06
buildwithllm.dev
This public portfolio: cinematic story, project ledger, transparent stack, and privacy-first contact.
Stack: HTML, CSS, JavaScript, Caddy, Oracle Cloud VM, HTTPS. Current: live and monitored over HTTPS.
05 / tools & languages
The tools are chosen for the job.
Fast prototypes, careful boundaries, and systems that can be understood by the people who inherit them.
Build
Interfaces that earn their place.
HTML, CSS, JavaScript, TypeScript, React, Next.js, Vite, Tailwind, Framer Motion, Three.js and React Flow.
AI systems
Models inside real workflows.
OpenAI Codex CLI, Claude Code, Gemini CLI, Hermes, MCP, Ollama and OpenAI-compatible endpoints.
Data & runtime
Small, observable foundations.
Node.js, Python, Express, Socket.IO, Prisma, SQLite, Supabase/Postgres, RLS, Docker, Linux and WSL.
Proof & delivery
Make the claim verifiable.
Graphify, Playwright, Vitest, ESLint, GitHub Actions, Caddy, HTTPS and Oracle Cloud Always Free infrastructure.
Graphify maps code, files, and relationships into a queryable project graph. It supports engineering decisions; it is not a substitute for human review.
05 / the working layer
Let the useful part come into view.
A good interface does not show everything at once. It reveals the right signal at the right moment, then gets out of the way.
- 01Quiet by default
- 02Clear when it matters
- 03Built for the next decision
Move across the surfaceThe working layer is always moving →
07 / approach
Clarity before complexity.
Every engagement begins with a concrete job to be done and ends with a next step that can be explained.
- 01
Frame the opportunity
Define the job, the people involved, and the signal that success is real.
- 02
Build the smallest useful version
Test the workflow in its real context before committing to a larger system.
- 03
Make it dependable
Document the boundaries, measure the result, and make the next iteration obvious.
08 / contact
The next useful system starts with one sharp question.
Bring a messy process, an early idea, or a question about what AI could do for a team.
Email EliasEmail is opened only when you choose the link. This site has no analytics, advertising cookies, newsletter, or automated visitor profiling. See privacy and legal.