In Conversation with Arash Pendari, Founder/Creative Director at Vionlabs

A Vitrina LeaderSpeak Podcast Summary

Vitrina: How would you describe what Vionlabs does?

Arash Pendari:
We’ve built AI models purpose-built for the media and entertainment industry. Our goal is to scale creative and repetitive tasks—like generating thumbnails, short clips, and metadata—so content platforms can operate like Netflix. What makes us unique is the focus on domain-specific models, not general-purpose ones. For instance, we generate thousands of creative assets per title so product and marketing teams can personalize the experience by audience demographic or context. We democratize this capability for platforms beyond the top-tier players.

Vitrina: What inspired you to start Vionlabs, and what expertise does your team bring?

Arash Pendari:
I studied VFX and game design—things like particle effects, water simulations, and working in Maya and Unreal Engine. But I say I’m a failed filmmaker who ended up building the tech behind the camera. We founded Vionlabs in 2010 and started working with AI around 2017-18, before TensorFlow was even standard. Our team includes machine learning experts and engineers trained to understand how humans emotionally perceive content—visuals, audio, narrative. That’s a complex challenge. Our experience in storytelling and emotional analysis gives us an edge.

Vitrina: Where exactly in the entertainment supply chain does Vionlabs operate?

Arash Pendari:
We stay out of post-production and VFX. Instead, we focus on user experience, media operations, and marketing. For example, Netflix revealed they use over 1,000 assets per title—thumbnails, logos, short clips, trailers—to enable personalization. Most streamers don’t have that. We help them generate these assets at scale. We also support media ops teams by automating tasks like detecting intros, credits, and episode recaps, which are tedious to do manually across long-running series like anime or telenovelas.

Vitrina: Who are your clients, and what are their common needs?

Arash Pendari:
We work with clients like Hulu, Paramount, Plex TV, and ProSieben, as well as telcos like Deutsche Telekom. Anyone managing a large video library—especially those struggling to package and surface content for engagement—needs what we offer. For example, one client filters joyful, romantic scenes to build thematic carousels for Valentine’s Day campaigns. Another organizes thousands of titles for internal editorial use or targeted ad packaging. The larger the archive, the more value we unlock.

Vitrina: How does your AI process video content and generate metadata?

Arash Pendari:
Our foundation is a 1,688-dimensional video embedding—a kind of emotional fingerprint. It clusters content by emotional similarity, so a scene from Skyfall might cluster with other spy thrillers. The AI detects mood (using valence, arousal, dominance psychology models), classifies keywords from transcripts (e.g., “MI6,” “James Bond”), and analyzes audio for stress and sentiment. For instance, it detects a sniper scene in Skyfall as fearful and suspenseful—automatically labeling it as such for clip generation or ad placement.

Vitrina: How do you handle international and non-Western content?

Arash Pendari:
When we started working with Hulu Japan, we had to train the model to understand anime differently from Western animation. We built feedback loops and retrained for Japanese, Middle Eastern, and Bollywood content. Bollywood, for example, can be visually joyful even during dramatic scenes. We taught the AI to interpret context accordingly. Now we handle most global formats. Also, since storytelling is becoming globally inspired, the emotional signals often align across cultures.

Vitrina: What’s the role of AI in improving personalization and recommendations?

Arash Pendari:
Personalization starts at the scene level. We extract emotional data per clip—joyful, tense, romantic—and match it to user preferences. Instead of waiting for a viewer to watch hours of content to figure out their taste, we can show them a suspenseful 10-second clip based on what we already know. We even personalize thumbnails: if one user skips Pulp Fiction when seeing John Travolta, we may swap in a poster featuring Uma Thurman next time. It’s what I call “packaging by personality.”

Vitrina: What about ad placement and monetization?

Arash Pendari:
We use multiple neural networks to identify optimal ad break locations. If all networks agree, it’s ranked as a perfect break—typically at a scene change with no dialogue or music. We also avoid inappropriate matches—like placing a car ad during a crash scene. We match IAB taxonomy, mood, and context to ensure ads enhance, not interrupt, the viewer experience. In fact, a U.S. survey we conducted showed poor ad experiences are a top reason people abandon AVOD services.

Vitrina: How is Vionlabs deployed, and what’s the typical integration process?

Arash Pendari:
Our software is deployed on-prem—inside the client’s own cloud infrastructure, like AWS or Google Cloud. Nothing is uploaded externally. We use Terraform scripts to spin up the infrastructure and can go live in as little as one to two weeks depending on client resources. Our API returns all scene-level data, mood tags, embeddings, and thumbnails in one call. It’s designed to integrate with recommendation engines, CMS systems, and marketing tools without disrupting existing workflows.

Vitrina: What’s next for Vionlabs in terms of innovation and market expansion?

Arash Pendari:
We’re focused on improving what we already have: better thumbnail logic, celebrity detection, brand recognition—especially for e-commerce and product placement. But we prioritize based on what our clients can use now vs. later. Geographically, we’re expanding across Asia, building on our presence in Japan, and exploring partnerships in the Middle East. We’re still a 28-strong team, but tech-heavy, and always aiming to solve problems that actually improve the creative process without replacing it.