Case Study: ARTBAT LIVE
AI-Driven Artist Discovery via Semantic Search
When ARTBAT LIVE’s manual tagging system stopped scaling with its artist community, Rosenblatt delivered the core of an AI-powered discovery MVP: a custom CLIP-based semantic search engine that matches clients to artists from a text description or an uploaded image. We provided the senior technical direction for ARTBAT’s development team in Hong Kong, validated the founder’s ideas with rapid proofs of concept before committing to a full build, and established the KPIs and North Star metric that kept the work pointed at outcomes.
“It has been a pleasure to work with the Rosenblatt Team. Their deep knowledge of AI, combined with their ability to adapt and tailor solutions to our specific use case, has been invaluable. The team’s expertise and flexibility have significantly contributed to laying a solid foundation for developing our application” - Mike Ha, CEO of ARTBAT LIVE
The Challenge
ARTBAT LIVE is a hybrid “Digital Art + Esports” platform hosting live 20-minute drawing battles, built around a core mission of artist empowerment: helping artists build portfolios, connect with clients, and turn their work into merchandise. As the artist community grew, manual tagging became a bottleneck on both sides of the marketplace; artists lost time keyword-tagging their work, and clients were limited to exact keyword matches that routinely missed relevant artwork. The scaling problem hid a second challenge: ARTBAT’s development team in Hong Kong needed senior architectural direction and prioritization to keep the build aligned with the CEO’s goals.
How We Worked
We worked backwards from the outcomes ARTBAT wanted to the tools that enable them; the fine-tuned CLIP model and full-stack AWS infrastructure were selections from a toolbox, not the starting point. Before committing to a full MVP, we used rapid proofs of concept to test whether the founder’s ideas were technically feasible; ideas that proved out were promoted to the roadmap, and ideas that didn’t were killed cheaply. We established KPIs and a North Star metric so that everyone, from the founder to the delivery team, shared one answer to the question that matters for any AI product: is the system improving outcomes or not? And as ARTBAT’s forward-deployed product engineers, we owned those outcomes; that included making the architectural and prioritization calls for the Hong Kong team and enforcing the policies needed to hit them.
What Shipped
Clients can now search by natural-language description (“a brutalist style with two men standing in front of a doorway”) or by uploading a reference image and get matched to artists based on the actual visual qualities of their work. Artists simply upload their portfolios; the system understands the work automatically. The MVP also shipped portfolio-level search for clients who want an artist with a history of similar work, and an Up & Coming Artists feature surfacing recent uploads on the landing page.
Full architecture details, including why we chose a fine-tuned CLIP model over an LLM, are available in the ARTBAT Technical Report.
Why This Matters for You
If you have a strong product vision but need senior technical judgment on architecture and prioritization, that’s the gap Rosenblatt’s forward-deployed product engineers fill. We ship alongside your team, de-risk ideas with fast proofs of concept, and instrument your AI product so you know whether it’s working.
If that sounds like your situation, click below to schedule a 30-minute meeting with our CEO and founder.