Datology Curation Studio launches · Launch Video Breakdown: Hook, Pacing & Motion Design

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:25

   The Hook

   Introducing Datology Curation Studio

   “Today, we're excited to launch the Datology Curation Studio, customizable frontier data curation for every team training its own models. Data quality is the ultimate compute multiplier, and Curation is the ultimate data quality lever. We're building Datology to be the AI training data refinery. We take raw, low-quality data and transform it into high-quality, high-octane data, enabling your models to learn more effectively. Training AI models is hard. It's expensive, it's slow, and it's often frustrating. The biggest bottleneck is data quality. Low-quality data leads to poor model performance, wasted compute, and missed opportunities. Datology solves this by improving data quality, reducing computational costs, and accelerating model development.”

   On screen

   datology.ai Ari Morcos CEO & Co-founder

   Camera

   Medium shot of the founder speaking directly to the camera, with occasional cuts to a wider shot of an office environment and data visualizations.

   Motion

   Direct address with supporting B-roll and animated data graphics. Text overlays introduce the speaker.

2. 0:25 – 1:00

   Problem Agitation

   The Data Quality Bottleneck

   “We're excited to introduce the Datology Curation Studio. The Curation Studio is a powerful platform that allows you to process, clean, and curate your data within your own private cloud. This ensures that you maintain full control and ownership of your data. The studio supports a wide range of data types, including text, code, images, and video. It provides a suite of tools for data cleaning, deduplication, and filtering. You can also use the studio to generate synthetic data, augment your existing datasets, and optimize your training pipelines. The Curation Studio is designed to be flexible and scalable, allowing you to handle even the largest and most complex datasets.”

   On screen

   Thomson Reuters Expert Human Preferences: Thomson-1.0-Large vs External Models Win / Tie / Loss share of comparisons Combined vs GPT5.5 62% 10% 29% vs Opus 4.8 58% 13% 30% vs GPT 5.6 Terra 57% 12% 32% vs Sonnet 5 53% 17% 30% vs GPT 5.6 Sol 54% 13% 33% Legal vs GPT5.5 63% 8% 29% vs Opus 4.8 63% 10% 26% vs GPT 5.6 Terra 64% 9% 28% vs Sonnet 5 54% 16% 30% vs GPT 5.6 Sol 57% 10% 33% General vs GPT5.5 59% 13% 27% vs Opus 4.8 46% 17% 37% vs GPT 5.6 Terra 43% 18% 40% vs Sonnet 5 50% 19% 31% vs GPT 5.6 Sol 47% 19% 33% Note: ties are shown explicitly rather than folded into the win rate. Thomson-1 wins Tie Opponent wins YOUR VPC Public curation studio Your data YOUR STORAGE Staged training datasets READY TO TRAIN curation studio 2 Clean Connect your data. Set your targets. Select source datasets and targets Clean Curate Create Compose 01 / 04

   Camera

   Alternates between founder speaking and animated UI/UX demonstrations of the Curation Studio. Data visualizations are presented as full-screen graphics.

   Motion

   Animated data charts and UI walkthroughs are integrated with the founder's narrative. Smooth transitions between live-action and motion graphics.

3. 1:00 – 1:30

   Product Reveal

   Datology Curation Studio in Action

   “The Datology Curation Studio is designed to be a comprehensive solution for all your data curation needs. It allows you to clean, curate, create, and compose your datasets with unparalleled efficiency. With the studio, you can easily identify and remove noisy data, fill in gaps with synthetic data, and optimize your datasets for specific training objectives. Our goal is to empower every team to build powerful models with significantly less compute, making frontier-quality intelligence accessible to all enterprises.”

   On screen

   curation studio 4. Create gaps with signal. Set the synthetic budget Compose Clean Curate Create 01 / 04 39 strongest publicly curated datasets Selected with deep data expertise. DATASET EXAMPLES Nvidia Nemotron pretraining Web Nvidia Nemotron SFT Web Math Code Stack v3 Code FineWeb Web FineWeb2 Multilingual Dataset Training Cycle 01 BASELINE Select best public data 02 CURATE Curate with Datology 03 TRAIN Train matched models 04 POST-TRAIN Verify after SFT + RLVR THE PROOF Benchmark performance Baseline Curated Mean benchmark score (%) 50 45 40 39.1 35 30 29.0 25 22.9 20 1.47 x 10^21 7.51 x 10^21 3.73 x 10^22 Pretraining compute (FLOPs, log scale) Source: Datology LLM technical report

   Camera

   Focus on animated UI/UX, data flow diagrams, and benchmark graphs. Founder appears briefly to reinforce key points.

   Motion

   Complex data visualizations and UI interactions are animated with a clean, modern aesthetic. Data points and lines animate to illustrate performance gains.

4. 1:30 – 1:54

   Call to Action

   Unlock Frontier-Quality Intelligence

   “Datology's approach allows for building powerful models with significantly less compute, making frontier-quality intelligence accessible to all enterprises. Join us in shaping the future of AI.”

   On screen

   curation studio datology.ai Trusted by Thomson Reuters HRT Deepgram [un]conventional ai Arcee

   Camera

   Returns to the founder speaking, followed by a full-screen logo reveal and partner logos. The camera maintains a medium shot of the founder.

   Motion

   Direct address, followed by a clean logo animation and partner showcase. Subtle camera zoom on the founder for emphasis.