AI experts from Meta, Databricks, Modal and more speak at Air Street Capital's NYC AI meetup hosted by Celonis

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:819390ca-bae1-46dd-b510-203fe805812c/as/231219_Celonis-for-AI-Explainer-video_Preview-Image.jpg

Artificial intelligence is reshaping the business world at an unprecedented pace. From automation to analytics, CRM to R&D, AI offers immense potential for driving innovation, optimizing operations, and uncovering valuable insights.

In May, researchers, developers and entrepreneurs from around the Big Apple gathered at Celonis’ New York City headquarters to hear from distinguished AI experts during Air Street Capital’s NYC AI Meetup. The event was a platform for lively discussions, knowledge sharing, and networking. It was a chance for those in attendance to gain insights into the latest advancements, challenges, and practical applications of AI.

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:097b201d-d7f4-4008-8c46-52b168786cc4/as/celonis_air_street_nyc_ai_meetup_nathan_air_street.png

Nathan Benaich (left) kicks off the Air Street Capital NYC AI Meetup at the Celonis New York City headquarters.

The speaker lineup included:

  • Nathan Benaich, founder and general partner, Air Street Capital
  • Eugenio Cassiano, SVP strategy and innovation, Celonis
  • Marc Kinast, VP corporate development, Celonis
  • Cody Blakeney, research scientist, Databricks Mosaic AI
  • Roshan Rao, research scientist, AI stealth startup
  • Akshat Bubna, co-founder and CTO, Modal
  • Laurens van der Maaten, distinguished research scientist, Meta AI

The research and applied AI talks covered a wide range of topics, including advances in generative AI models, LLM training methodologies, using AI for applications such as predicting protein structures, and how process intelligence is key to unlocking the full potential of enterprise AI.

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:53755e38-ea90-4278-bc0d-cee0673b191d/as/celonis_air_street_nyc_ai_meetup_marc_eugenio.png

Marc Kinast (left) and Eugenio Cassiano (right) walk attendees through the Celonis Process Intelligence Platform.

Eugenio Cassiano and Marc Kinast got things started by outlining Celonis' and the company’s Process Intelligence Platform. Process Intelligence (PI) provides the critical layer of business context that closes the gap between raw business data and artificial intelligence. PI enables AI to understand the complexities of how organizations actually operate across their systems, processes, and departments. Without this context, AI solutions remain siloed and disconnected. Cassiano used the example of an invoice to illustrate process intelligence’s importance.

“Take a normal enterprise invoice,” said Cassiano. “A classic data model can tell you if it’s been paid on time or not, but to go further and understand connections, like the relationship between on-time payments and customer satisfaction, you need business context. Celonis provides this context, closing the gaps within your data and enabling better enterprise AI.”

Celonis process intelligence understands both the “as is” and “to be” states of a process and the difference between the two, it acts as the foundation of and fuel for AI and automation across business operations. Celonis also solves typical LLM issues, such as hallucination and data staleness.

Cassiano also explained how Celonis has continuously evolved its underlying models, and embedded AI throughout its platform, such as with Process Copilots. Using the Celonis ML Workbench and LLM templates, customers can craft bespoke AI solutions. And with the company’s Intelligence API, they can feed process insights to AI anywhere in their tech stack.

Read: How industries are pairing AI and process mining to drive transformation

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:4608400b-9014-412e-8bf1-3a2dc41b9af3/as/celonis_air_street_nyc_ai_meetup_cody_databricks.png

Cody Blakeney talks about Databricks and Mosaic AI’s DBRX LLM.

Cody Blakeney was next to speak and walked the audience through the methodology the Databricks Mosaic AI team used to train DBRX, an open, general-purpose large language model (LLM) released in March. He also shared learnings from their work, including his advice to “make fewer bets, but make them bigger.”

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:566b0df6-b696-4ce7-a5ee-c5e94a74b097/as/celonis_air_street_nyc_ai_meetup_roshan_startup_stealth.png

Roshan Rao talks about using AI to create new protein molecules.

Switching from the purely digital to the biological, Roshan Rao shared insights into the work his company (a startup still in stealth) is doing to train AI models on protein sequences with the goal of designing new therapeutics.

“Antibodies are a type of protein,” said Rao. “If you can model them really well, and you can predict what they will fold into and what they will bind to, whether they will target a specific spike protein, whether they will target a receptor on a cancer cell, and you can predict how with high accuracy, then you can design new antibodies that will be therapeutically useful.”

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:4a82fbe3-f411-4e71-841d-07cd476db439/as/celonis_air_street_nyc_ai_meetup_akshat_modal.png

Akshat Bubna (left) demonstrates Modal.

Akshat Bubna gave attendees a run-through and live demonstration of Modal, a serverless platform that uses a custom container system to run code in the cloud. The platform can be used for job queues and batch processing (like running Cron jobs), but as Bubna demonstrated, it can also provide the infrastructure for AI model inference and fine-tuning work.

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:95130fa4-b118-42a4-a9ae-018fce07ba50/as/celonis_air_street_nyc_ai_meetup_laurens_meta.png

Laurens van der Maaten talks about Meta Llama 3

Last to speak was Laurens van der Maaten with Meta AI, who provided a walkthrough of the work his team did on Llama 3, Meta’s openly available LLM released on April 18. He discussed their approach to scaling up the model training, data filtering, quality control, and expanding Llama to multilingual and multimodal capabilities, such as the recently released Ray-Ban Meta smart glasses. He also affirmed Meta’s commitment to safety and an open AI ecosystem.