Rohit Prasad

Apr 27, 2026

• "we are giving multiple checkpoints, the pre-trained checkpoint, the mid-train checkpoints, the post-train checkpoints. At which of each of these model training stages, you can add your frontier data, your proprietary data"

  cited in Developer · from AWS re:Invent 2025 | Rohit Prasad, AWS

• "There's a persistent challenge. A frontier model comes out. Public benchmarks look great. Internal and external customers, that was happening to us internally as well. People try it. And then the production reality sets in that when you try to build your applications, your workflows, it doesn't meet your expectations"

  cited in Deep Tech · from AWS re:Invent 2025 | Rohit Prasad, AWS

• "we are giving you access to Amazon curated data to blend in. And then when you do that, the model maintains its general intelligence and becomes an expert in your domain."

  cited in Deep Tech · from AWS re:Invent 2025 | Rohit Prasad, AWS