01
Diagnose
Map the existing system. Find what's missing.
- Interviews
- Audit
- Data
AI Systems Engineer at Raving Fans
I build systems that think.
I connect strategy, automation and AI so organizations run faster, without the overhead.
Manual work in.Finished systems out.
A week of bookings turns into sent invoices, without anyone touching a spreadsheet.
Illustration of an operations console running the weekly billing pipeline. Twelve bookings are fetched, matched with their creative files, numbered and sent as invoices. One booking has no creative file, so it is held and the ops team is notified by email. Result: 11 invoices sent, 1 held, 0 manual steps.
About me.The person behind the systems.
Frankfurt
I'm Manu. 25. AI Systems Engineer at Raving Fans, working remotely from Frankfurt.
I spent years trying things. Dropshipping, copywriting, product management, course creation. Every time I hit a ceiling, I'd look for a smarter approach. When I started using AI to systematically break down problems, not just in business but personally, something clicked: you can build almost anything if you think carefully about what actually matters.
What drives me isn't the end product. It's the full process. Understanding the goal, mapping the current state, deciding what's worth solving and where you need a workaround. Good systems take work off people's plates. That's what I care about.
Outside of work: gym, meditation, and traveling. Usually somewhere in Asia.
What I build.Systems, not slide decks.
Three kinds of work, usually inside the same project. Each one ends in something that runs in production, not in a prototype.
Automation systems
End-to-end workflows that eliminate manual work. From API connection to finished process.
- Make.com
- n8n
- Zapier
- Webhooks
AI integration
OpenAI and other models used purposefully. Not as toys, but as leverage.
- OpenAI
- Claude
- RAG
- Prompt design
Full-stack products
Internal tools, dashboards, and platforms that people actually use.
- Next.js
- Supabase
- TypeScript
- Vercel
How I work.Four steps, every project.
The same loop, whether it is a two-day automation or a platform with twelve modules.
- 02
Design
Define the simplest architecture.
- Flowchart
- Scope
- KPIs
- 03
Build
Connect the tools. Ship fast.
- Code
- Wire
- Test
- 04
Iterate
Measure output, cut waste, repeat.
- Metrics
- Trim
- Loop
Selected work.Systems that run every day.
Platforms, automations and AI agents, built end to end and in daily use. Filter by type, or watch a few of them run in the lab below.
Internal Operations Platform
Operations ran on spreadsheets and disconnected tools that didn't talk to each other. Built one platform instead: CRM, project management, time and vacation tracking, OKRs, and a client-facing portal in a single app. 12 modules, architecture to deployment, solo. Runs the entire company, every day.
- Screens
- 26
- Tests
- 437
- Daily use
- Live
- Next.js
- TypeScript
- Supabase
- Railway
- shadcn/ui
Lead Pre-Qualification Workflow
A practice in the health sector sorted every enquiry by hand: course access, a questionnaire, a decision, two CRMs and the right reply. Built three n8n workflows that do all of it: a five-question decision tree routes each case to one of seven outcomes, both CRMs update without duplicates, and the matching WhatsApp message waits for the third working day. All 64 answer combinations are covered by tests.
- Nodes
- 83
- Integrations
- 8
- Paths tested
- 64/64
- n8n
- JavaScript
- JotForm
- Close
- Superchat
AI Creative Engine for Webinar Ads
A webinar-marketing business needed dozens of on-brand ads for every launch, and generic AI images kept breaking the brand. Built a creative engine from three services: a chat with copy, creative and QA agents, plus two MCP servers that hold brand values, real photos and a deterministic renderer for 13 ad formats. AI writes and picks, code lays out, and a vision check flags whatever slipped.
- Tests
- 889
- Ad formats
- 13
- Services
- 3
- Python
- MCP
- Anthropic API
- LibreChat
- Railway
AI Support Agent with Learning Loop
The support team of an online education business answered the same questions every day from memory. Built an agent that reads each new ticket, finds the five most similar past cases and writes a draft that a human sends. A learning loop compares every draft with the answer that actually went out, feeds good corrections back into the knowledge base and tracks the score week by week.
- Tests
- 181
- Services
- 6
- Score report
- Weekly
- Python
- Anthropic API
- Supabase pgvector
- Railway
- MCP
Automated Media Billing Pipeline
Weekly invoicing for a media-buying client was a manual grind: check bookings, match creative files, number invoices by hand. Built a pipeline that does all of it and escalates exceptions by email instead of failing silently. Runs unattended, every week.
- Tests
- 372
- Manual steps
- 0
- Fully automatic
- Weekly
- Python
- Airtable API
- SharePoint
Lead Scoring & Funnel Platform
A tenant-representation service needed to separate hot leads from noise without hiring more sales staff. Built a 5-step funnel with a server-side scoring engine and an authenticated admin dashboard for lead management. 59 tests across 12 suites, shipped in three days.
- Tests
- 59
- Days
- 3
- Step funnel
- 5
- Next.js
- TypeScript
- Supabase
- Jest
- Vercel
Outbound Lead Engine
A hotel group wanted more meeting and event bookings from companies nearby, without a sales team. Built a self-hosted engine that finds companies around each hotel, rates their fit with an LLM, verifies emails and runs a three-step sequence that stops the moment someone replies. Every reply comes with an AI draft, and safety rails decide what may go out automatically.
- Tests
- 113
- Step sequence
- 3
- Reply detection
- Auto
- Python
- FastAPI
- Anthropic API
- Airtable
- IMAP
Affiliate Referral Platform
No system existed to track referrals or pay commissions across multiple tenants. Built a multi-tenant platform from scratch: three permission levels, EU-hosted, real-time dashboards, and commission logic built in.
- Tenants
- 3
- Weeks
- 11
- Roles
- ×3
- React
- TypeScript
- Supabase
- Vercel
- Edge Functions
AI Health & Training Coach
Sleep, training and calendar lived in three apps that never talked to each other. Built an agent loop without a framework behind one Telegram persona with long-term memory, fed by Oura recovery data, the training log and the calendar. The math stays in code and the model only writes: morning check-ins, post-workout reviews and a Sunday report.
- Tests
- 71
- Data sources
- 3
- Check-ins
- Daily
- Python
- SQLite
- Telegram Bot API
- Oura API
- Hevy API
Radio Spot Production App
Radio orders were handed to copywriters and voice talents through spreadsheets and email threads. Built a web app where dispatch assigns each order, copywriters submit script and direction, voice talents upload the take, and every change is logged by person and field. The finished spot flows straight into the billing pipeline.
- Tests
- 69
- Pages
- 21
- Change log
- ✓
- Next.js
- TypeScript
- Supabase
- Airtable
- Microsoft Graph
Customer Feedback Loop
Feedback surveys at a media sales company went out late or not at all. Built a loop between Airtable, a small approval app and n8n: paid and approved customers show up every working morning, staff approve with one click, n8n sends the right survey and writes the result back. Anyone silent after three working days gets exactly one reminder.
- Click to send
- 1
- Working days to reminder
- 3
- Survey stages
- 2
- n8n
- Airtable
- Fillout
- Node.js
Don't take my word for it.Run the systems yourself.
Three of the systems above, rebuilt as small simulations. Nothing here talks to a server: the logic runs in your browser, the same way it runs in production, just faster.
01Billing pipeline
Weekly invoice run
The unattended Monday job from Selected Work. Flip the switch to see how it handles a broken booking instead of failing silently.
02Lead scoring engine
Score a lead
The server-side scoring from the funnel platform, exposed. Change any answer and watch the score, the tier and the routing decision update.
03Agent loop
Weekly health report
The Sunday run of the health coach. Pick a week, watch the loop gather data, do the math in code and hand the model only the part a model is good at: writing.
How I got here.Consulting, marketing, then systems.
Six years, three industries, one thread: finding where processes break and building something that holds.
2020–2022
Accenture, IT consulting
First real job. Dual study in IT consulting.
Learned how large organizations run, and where they break.2022
LG Electronics, D2C marketing
Back-office in direct-to-consumer marketing.
Understood funnels before I knew what funnels were.2022–2024
TH Aschaffenburg, South Korea and Asia
Finished my BBA. Semester abroad in South Korea.
Japan, Vietnam, Thailand, Taiwan, Singapore, Malaysia.
Started thinking about what I actually wanted to build.2023–2026
6FE, course infrastructure and AI systems
Designed module architecture, AI writing workflows, and automation systems for a freelancing coaching program actively serving the LATAM market.
2026 →Now
Raving Fans, AI Systems Engineer
Joined in February. Built the internal ops platform (12 modules), an affiliate platform and an automated weekly invoicing pipeline, plus systems for clients: n8n workflows, a support agent with a learning loop and an AI creative engine. Architecture to deployment. All of it.
Got a process that should run itself?Let's build it.
I'm open to interesting problems. Tell me where your team loses time every week.
Email me
hi@manuelrocholl.com