Visual Layer was a computer vision AI startup working with very large image and video datasets. I was Head of Marketing. In February 2026, we held the company's first technical meetup at our Tel Aviv office, with talks from CTO Amir Alush, Senior Backend Engineer Amir Baron, and Algo Research Lead Guy Singer.
The audience was deliberately technical: engineers and data architects, AI researchers, product owners working on search, and technology leaders building RAG or natural-language systems over visual data. The program covered temporal video search, architecture at petabyte scale, Visual Layer Chat, agentic workflows, production failure modes, and natural-language queries over visual data.
The technical problem
One engineering problem gave us a useful subject for the event.
Large visual datasets produce huge amounts of metadata: labels, tags, objects, bounding boxes, classifications, and other information attached to the underlying images and video. Passing all of that information directly into an LLM consumes the context window very quickly. Visual Layer had developed an architecture that let the model work with references to the data, while other components handled retrieval and the underlying visual information.
Building the event around engineering
I worked from the engineering material outward.
The broad event description started with search over video and visual data. The agenda then went deeper into temporal search, visual graphs, agents, system architecture, and production problems. As we promoted the meetup, the content became more specific: metadata cardinality, context-window limits, pass-by-reference architecture, and how agentic workflows operate over visual data.
The speakers' promotion used the same vocabulary. Amir Alush described architecture tradeoffs at petabyte scale, failure modes, agentic workflows, and lessons from building Visual Layer Chat.
I also produced a short animated teaser for the meetup using Remotion, Claude, and VS Code. I planned the content and assets, then worked through the animation, timing, and visual hierarchy until the video could explain the architecture rather than simply advertise the event.
Alongside the public campaign, I handled the operational communication: internal promotion, employee sharing, direct attendee outreach, and RSVP confirmation so we could plan the room, food, and drinks against the people who actually intended to come.
Why I built it this way
The people we wanted already worked with LLMs, search systems, computer vision, and data infrastructure. They needed enough technical detail to decide whether the session deserved their time.
So I kept the marketing close to the actual system design. The event page named the problem and the intended audience. The technical posts explained the architecture. We also launched the event via Meetup at the same time in order to track registration of course. The speaker posts covered the material they planned to discuss. The video made one difficult part of the system easier to see.
Across all of them, I used the same technical terms for the same concepts. That matters to me in technical marketing because changing the vocabulary between engineering material, event promotion, and product messaging makes the reader do extra translation.
