Open by Design: NYC Developer Meetup will bring Together AI, NVIDIA and Reflection AI into a discussion of open models in production on October 14, 2026, in New York City. The evening runs from 6–9 PM EDT, with host approval required for registration. Its premise reaches a consequential engineering decision: how to move from a model that works in an experiment to a system a business can operate.
For engineers, researchers and technical founders, that decision carries more than a model preference. Evaluation, customization and serving all affect what a team can build and what it will have to maintain. The gathering offers a place to examine those choices with people working across developer experience, reinforcement learning and developer relations, before another production decision becomes somebody's weekend.
Open by Design NYC centers on evaluating production workloads, moving from closed to open models, and the role of post-training and fine-tuning. That agenda puts responsibility alongside flexibility. Access to a model creates options, while the application team still has to decide which options fit the job. A useful discussion begins with the workload: the inputs it receives, the failures it produces and the behavior a customer needs.
Together AI's evaluation platform provides a concrete example of this direction. In its February 2026 announcement, Together AI describes a repeatable way to compare proprietary, open and fine-tuned models. The point for an operator is the evaluation method, rather than a universal winner. A company choosing a model for a specific task needs evidence that speaks to that task, with enough consistency to compare the next candidate fairly. DevCuration's coverage of AI evaluation in production traces follows the adjacent question of making model quality visible inside a running application.
An impressive answer in a demo can leave several practical questions untouched. How does the system behave when the input is incomplete, when a tool fails or when several requests arrive together? Those are questions attendees can bring to the room, rather than promised agenda items. They make the business stakes tangible: a model decision becomes a release decision, and a release decision eventually reaches the person accountable for the customer's experience.
Together AI's fine-tuning announcement describes preference optimization and continued training as ways to adapt models to application needs. That gives the meetup's customization theme a practical foundation. The source explains the available tools; the decision to use them still belongs to the team. A founder considering customization should be able to explain the behavior being improved and how an improvement will be recognized.
NVIDIA's NIM developer documentation describes containers for self-hosted inference across clouds, data centers and workstations. Reflection's official site describes open-weight models, open-source software and deployment options as parts of its stack. Together, these vendor descriptions show why an open-model conversation can quickly reach serving, infrastructure and control. They supply context for the organizations in the room, rather than evidence that a particular deployment will outperform another.
For a technical operator, the choice has an ownership dimension. Customization can create work around data selection, evaluation and the next model update; deployment decisions assign responsibility for running the resulting system. An attendee could use the fireside to ask how those responsibilities are divided and where the expensive surprises tend to appear. A clear account of the work can be more useful than another claim that a model is ready for production.
The organizer lists Hassan El Mghari, Head of Developer Experience at Together AI; Zach Wentz, Member of Technical Staff working in RL at Reflection; and Miriam Lebowitz in Developer Relations at NVIDIA. Those are the roles given for this event. The combination puts application-facing work, model research and developer adoption into one discussion, where each perspective can expose a different assumption about what a production team needs.
A developer experience perspective can make integration and day-to-day usability part of the conversation. A reinforcement learning perspective can bring attention to how model behavior is shaped and evaluated. Developer relations can connect technical capabilities to the questions builders ask when they try to use them. These are reasons the listed roles are relevant, not predictions about specific remarks or a substitute for an announced talk abstract.
Together AI Calendar presents the event, while Together AI and NVIDIA appear as the explicit host accounts. Reflection AI is named in the description as an involved organization. The proposed fireside format gives the gathering a focused center, followed by time for conversations among attendees. A builder carrying a concrete migration problem could find that sequence useful: listen to the different perspectives, then ask a narrower question of someone dealing with a similar constraint.
Open by Design's proposed agenda starts with doors, drinks and networking at 6 PM. The fireside conversation is listed for 7 PM, followed by drinks, bites and networking at 7:45 PM, with the evening concluding at 9 PM. New York City is the confirmed location; the exact venue and street address are available through registration. Prospective attendees should use the official listing to request access and confirm the details after approval.
The intended audience includes engineers, researchers and builders working on open models. Technical founders and operators weighing a migration have a practical reason to join that conversation with a specific use case. They could leave with better questions about model selection, customization or serving, and connections to people working on adjacent problems. The value depends on the discussion and the people they meet, rather than any promised customer, investment or career outcome.
New York's AI gatherings can put quite different forms of judgment under discussion. DevCuration's OPEN TALKS coverage examines a separate October 14 conversation about generative AI and creativity. Open by Design takes the production route into that wider question of what builders should trust. For the engineer evaluating an open model, the next conversation can start with the task that still fails, the evidence needed to change it and the person who will own the release.
Bring a specific workload and a question about evaluation, migration, fine-tuning or serving. The announced discussion centers on moving open models into production, so concrete application constraints give the conversation a useful starting point.
Registration is subject to host approval, and the organizer says space is limited. New York City is confirmed, while the exact address is provided through registration; check the official listing after approval.
Together AI Calendar presents the gathering, and Together AI and NVIDIA are the explicit listed host accounts. Reflection AI is named as an involved organization in the description and has a participant in the announced fireside lineup.
The organizer lists Hassan El Mghari in Developer Experience at Together AI, Zach Wentz in RL at Reflection, and Miriam Lebowitz in Developer Relations at NVIDIA. Those roles connect application building, model research and developer adoption; specific remarks or outcomes have not been announced.
The October 14, 2026 agenda is proposed: networking at 6 PM EDT, a fireside at 7 PM, further networking at 7:45 PM, and a 9 PM finish. Evaluation, post-training and fine-tuning are discussion topics, rather than a promised hands-on workshop or a guaranteed migration result.