I build AI agents that have to work for real users.
I study IT-Systems Engineering at HPI in Potsdam (GPA 1.44). At MitJulia I work on the voice agent that answers the phone in 100+ salons1. Until October I was Founding Engineer at Kobe AI2.
Four stations
The dots in the background change shape as you scroll: first a network, then a voice, a chest X-ray and the northern lights. Each station comes with one number you can check.
01 · Agents2026
Until October 2026 I was Founding Engineer at Kobe AI. Kobe searches your own network for the person who can introduce you, and sends nothing before you confirm. I worked on the Cloudflare Workers backend and the Electron client, and I extended the MCP server.
~470 commits in three full-time weeks, October 2026
02 · Voice2026
At MitJulia I work on Julia. She answers the phone in hair salons, checks the calendar and books the appointment. The stack is Twilio, ElevenLabs, AWS Lambda, DynamoDB and TypeScript. A simulator runs test calls against the agent and an LLM judge grades them.
100+ salons live on the voice agent, November 2026
03 · X-ray2024
PneuNet started at school. I wrote a convolutional network from scratch in NumPy to detect pneumonia on chest X-rays. Later I compared seven architectures in PyTorch and added lung segmentation and class activation maps.
>95 % segmentation on the project's own validation split
04 · North2026
The Arctic Sprint: 17 days in a campervan from Berlin through the Baltics to Rovaniemi and the Lofoten. We built the Kobe MVP on the road and drove straight on to START Hack in St. Gallen.
17 days, nine countries, one MVP
Signal / numbers
Where I have been
01 — Selected work
Finds the warmest path to the person you need, through your own network.
~470 commits in 3 weeks · Oct 20262
Ask Kobe
Your personal Superconnector
Talk to a payments lead at a European fintech
Introduce me to someone who leads payments at a European fintech. We want feedback on our checkout API before the pilot.
I found three warm paths to payments leaders. Lena is your strongest: she led Finlo’s seed round and you talk to her often. Review the plan and the drafts below.
Talk to a payments lead at a European fintech
The Plan
Ask Lena on WhatsApp to introduce you to Tobias, with a short message they can forward.
Ask David on LinkedIn to introduce you to Mara, with a short message they can forward.
1 more contact path included below.
After you approve, Kobe sends the messages and follows up to arrange your conversations. For introduction paths, Kobe waits until you are connected before contacting the target directly.
Approves all current message drafts in this Mission. Sends after a 30 second countdown.
You Through
LenaPartner · Northwind Capital To
TobiasHead of Payments · Finlo Lena led Finlo’s seed round. You have exchanged 14 messages with her since August.
Hi Lena, hope Lisbon was good! Quick ask: could you introduce me to Tobias at Finlo? We open our checkout API pilot in January and his view on European payment flows would help us a lot. I wrote a short note below that you can forward. Thank you!
Forwardable message
Hi Tobias, I’m Alex, building checkout infrastructure at Northbeam. Lena suggested you as the person to ask about payment flows in Europe. Would you have 20 minutes in the next two weeks? Happy to share what we learned from 40 merchant interviews.
Sent as a separate message immediately after your request.You Through
DavidFounder · Ledgerly To
MaraVP Payments · Paystream David and Mara worked together for four years. You met David at two founder dinners this autumn.
Hi David, good to see you at the dinner last week. Would you be open to introducing me to Mara? We’d love her take on our checkout API before the pilot. A short note to forward is below.
Forwardable message
Hi Mara, I’m Alex from Northbeam. David mentioned you know payments at scale better than anyone. Could we borrow 20 minutes of your time for feedback on our checkout API?
Sent as a separate message immediately after your request.You To
JonasPayments Product Lead · Cardwise Jonas is a first-degree connection and replied to you within a day last time.
Hi Jonas, it’s been a while since the API meetup. We’re opening a checkout API pilot in January and I’d value your feedback as someone who ships payments products. Up for a 20 minute call next week?
Sample people: invented names, stock portraits (Unsplash).
Sent. I’ll follow up with Lena and David on Thursday if they haven’t replied, and tell you as soon as an introduction is made.
MitJulia (formerly Friseurtelefon)
Julia picks up 24/7 and books straight into the salon’s booking system.
100+ salons · Nov 20261
Sat 7:42 pm · outside opening hours
Julia is listening
Init webhook · client recognised: Sophie
Hi Sophie, this is Julia, the digital receptionist at Sample Salon. How can I help?
Hi! I need an appointment next week, ideally Tuesday after work.
Sure. Wash, cut and blow-dry with Mia again?
Yes, exactly.
get-appointment-availabilities · Tue · Mia → 2 slots
On Tuesday Mia is free at 17:30 or at 18:30. Which suits you better?
18:30 is perfect.
create-booking → appointment-booked
Done. Your appointment is on Tuesday at 18:30 with Mia. See you then!
Thanks, bye!
Revenue
2,846 €
612 € outside opening hours
Bookings
47
11 outside opening hours
Calls
129Σ 2:41h · Ø 75s
31 outside opening hours
Calls over time
Call reasons
Results - call reason: Booking
| Time | Duration | Name | Call reason | Call result | Status | |
|---|---|---|---|---|---|---|
| Sat17:58 | 0:48 | Paul | Change | Appointment moved | Done | |
| Sat16:12 | 1:31 | Emma | Transfer | Transferred / callback | Open | |
| Sat14:40 | 0:36 | Unknown | Price | Information given | Done | |
| Sat11:05 | 1:04 | Leon | Booking | Appointment booked32 € | Done | |
| Sat10:20 | 0:52 | Hanna | Cancellation | Appointment cancelled | Done | |
| Fri18:47 | 1:20 | Amelie | Booking | No suitable slot | Open | |
| Fri16:30 | 0:41 | Noah | Callback request | Callback logged | Open | |
| Fri12:15 | 1:09 | Clara | Booking | Appointment booked89 € | Done |
Booking system · Tuesday
Julia books straight into the salon’s own booking system.
Appointment
30 teams from ~700 applications. I kept shipping Kobe and built the pitch deck in a day.
USD 100,000 · no equity
Result as announced to the teams; written confirmation pending.
Built in NumPy first, then seven architectures benchmarked in PyTorch.
Jugend forscht 2024: Thuringian state winner and a special prize at the national final.
> 87 % classification · > 95 % segmentation5
Arrow keys move the lens.
HPI seminar · industry partner BMW
An agentic system that spots early warning signs of unplanned machine stops.
Ongoing · team of six · under NDA
Machine events
Click a station to trigger a fault burst there
Pipeline (deterministic)
Detectstatistics on event patterns
Prioritiseimpact on the flow
Notifytell the right role
Learnfeed decisions back
Agent · works only through tools
Every number is a reference to a tool result
Draft · maintenance notification
Draft · nothing written yet
Concept illustration for the ongoing seminar “Trends and Concepts: Building Agentic Products” with BMW as industry partner. It shows only the design rules of the team’s concept: code computes every number, the language model works through tools and explains, a human approves every write-back. Stations, fault codes and values are invented; no partner data, no process details, not the team’s UI.
02 — Experience
MitJulia Software Engineer, AI Voice Agents · Julia answers salon phones 24/7 and books straight into the booking system. since Jan 2026 03 — Now
Team of eight at Prof. Gerard de Melo's AI chair: a multi-year pipeline that detects, tracks and re-identifies wild gorillas in camera-trap video.
I take an issue from the first report to the pull request, and check every salon on the way.
I plan the company visits, pick the participants and keep the budget.
Six of us build an early-warning system for unplanned machine stops. The details are under NDA.
04 — Awards
05 — Moments
06 — Blog
I wrote one post per project. Most of them come with a demo you can click through, filled with sample data. The other posts are about competitions, trips, volunteering and startup work.
For the Erasmus Student Network in Potsdam we mapped the event process, modelled it and shipped a small web app as a team. Understand first, then build.
In the seminar Building Agentic Products we're working with BMW on a system that spots early warning signs of unplanned machine stops. An interim update, as abstract as confidentiality requires.
Since October I've been in the bachelor's project that re-identifies wild gorillas in camera-trap video. An interim update on how you get into a pipeline that four cohorts built before you.
07 — Contact
Internship, project or a semester abroad: write to me. I usually answer within a day or two.
marvin.heyne@student.hpi.uni-potsdam.de
