Web applications
SaaS, internal portals, dashboards, headless e-commerce. Modern stack (Next.js, TypeScript, Postgres) shipped in sprints of weeks, not months.
A PERSONAL CONCEPT — AI AGENTS BUILDING SOFTWARE
A demo of how AI agents could ship custom software — with a human reviewing every line.
Web, mobile, back-office, chatbots, and AI products — imagined as a personal build experiment. Pipelines orchestrated by specialist agents (design, code, tests, deployment), with me reviewing every step.
Animated diagram: ten specialist agents (strategy, ux, frontend, backend, architect, security, seo, ci/cd, qa, monitor) feed an orchestrator that opens pull requests; a human reviewer approves each one before it ships.
Built with tools used in real production stacks
Next.js
TypeScript
Tailwind CSS
React Native
Expo
Kotlin
PWA
FastAPI
Python
PostgreSQL
Supabase
Upstash
Claude
- OpenAI
Vercel
Cloudflare
Google Cloud
Nginx
Docker
GitHub Actions
n8n
Sentry
Telegram
What the pipeline builds
Five capability areas. One agent pipeline. One human reviewing.
Mobile applications
Native iOS and Android, or React Native, with backend included. From MVP to production with CI/CD and observability from day one.Back-office automation
Integrations between ERP, CRM, spreadsheets, and APIs. Manual processes turned into reliable pipelines, with dashboards and alerts.Chatbots and conversational agents
WhatsApp, web, and voice. Bots that qualify leads, resolve support, and execute actions — with guardrails, evaluation, and human handoff.AI products and agents
RAG systems, autonomous workflows, evaluation harnesses. Hybrid Claude + local Ollama for cost and privacy control.
How a project like this would work
Four illustrative steps, each with a deliverable and a human sign-off before moving on.
- 01
Brief
One call and a document: use cases, constraints, success criteria — turned into a spec within 48 hours.
- 02
Design
Wireframes, schema, and architecture — then a clickable Claude Design prototype for sign-off, before a line of code is written.
- 03
Build
Agent pipeline (planner, coder, tester, reviewer). Small PRs, green CI, weekly demo.
- 04
Review and ship
Architect review (Opus), security audit, design-fidelity check, automated production health. Documented handover.
Stack and practice
Tools we use in production, not in slide decks.
- Next.js, React Native, FastAPI, Postgres, Cloudflare
- Agent pipelines on Claude Agent SDK + LangGraph
- CI/CD with GitHub Actions, deploy to Vercel, Cloudflare Workers, Railway, GCP, AWS, or fully on-prem
- Observability with OpenTelemetry, Sentry, and Phoenix
- Human review on every PR — code, security, and product
$ solaar pipeline run --project demo
[planner] spec parsed · 14 tasks
[designer] wireframes generated · awaiting human review
[coder] PR #3 opened · 0 lint errors
[tester] suite green · 42/42 passing
[reviewer] no high-severity findings
done | human checkpoints: 2/4 | elapsed: 3d 04h
Curious how a pipeline like this works?
This is a personal concept project — not a service for hire. Explore the case studies or say hi.