Key points
- Nandakishor Mukkunnoth asserts he designed Jev's core architecture a year before TypeSafe's release.
- TypeSafe’s Jev offers probability-based answers via API, avoiding standard LLM generation issues.
- Competitors and similar open-source projects have emerged quickly following the product launch.
TypeSafe’s Jev has faced allegations of copying prior research by ConvAI Innovations founder Nandakishor Mukkunnoth. The dispute centres on the timing and originality of the decision model architecture used in the hosted AI product.
What happened
TypeSafe released Jev, an API-driven decision model that provides probability outputs for specific questions, alongside a description of its System One architecture. According to Tom's Hardware, the release prompted a wave of similar projects from major industry players and open-source developers.
Nandakishor Mukkunnoth, founder of ConvAI Innovations, stated in a blog post that he designed the Jev architecture one year prior to TypeSafe's launch. He referenced a March 30, 2025, arXiv paper on SalesRLAgent, which detailed a model for probability estimation.
Mukkunnoth published his model weights and dataset alongside that initial paper. He later released a September paper that more closely aligned with the operational mechanics of Jev. He expressed frustration at being overlooked because his work targeted a vertical use case, whereas TypeSafe created a general-purpose decision engine.
TypeSafe delivered Jev as a complete hosted product, complete with servers, API access, documentation, and developer kits. The service charges users for access, distinguishing it from Mukkunnoth's open academic contributions.
Why it matters
The Jev architecture sidesteps common large language model issues by avoiding token-by-token text generation, conversational history management, and hallucination risks. It answers questions in parallel with direct probability data, making it faster and cheaper than standard LLMs for specific "if" statement logic.
The rapid emergence of competitors suggests high demand for this practical application, which allows developers to feed input data like shopping transactions alongside specific questions. Users receive yes/no or multiple-choice probabilities that integrate directly into their own code logic.
The attribution dispute highlights the tension between vertical research and general-purpose commercial products. Mukkunnoth’s claim implies that TypeSafe may have leveraged existing academic work without proper credit, raising concerns about intellectual property norms in the AI sector.
What happens next
- Monitor official statements from TypeSafe regarding the attribution claims.
- Review licensing terms for any open-source projects inspired by Jev.
- Evaluate the technical differences between Jev and earlier probability models.
- Watch for further publications or legal actions related to the architecture.
What to do and how to stay safe: TypeSafe
- Verify the provenance of AI models before integrating them into production systems.
- Review licensing agreements for open-source components used in decision engines.
- Document the origin of architectural designs used in commercial AI products.
- Monitor vendor communications for updates on attribution disputes or legal issues.
Step-by-step guide: Shadow AI usage: how to control unmanaged model access in your firm
General security guidance from the AppSec Scoop newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Who is claiming TypeSafe copied their research?
Nandakishor Mukkunnoth, founder of ConvAI Innovations, claims he designed the Jev architecture a year before its release.
What is the core function of Jev?
Jev is an API-driven decision model that provides probability-based answers to specific questions, avoiding standard LLM generation issues.
When was Mukkunnoth's related research published?
Mukkunnoth published his initial paper on SalesRLAgent on March 30, 2025, with a follow-up paper in September.
