Pelora — Generate Specialized AI Models

THE PROBLEM

Custom AI models are still too expensive and slow to create.

Organizations increasingly need models adapted to their own data, domains, and workflows. But today, creating a specialized model usually requires collecting data, running fine-tuning jobs, evaluating results, and repeating the process — often across multiple expensive iterations.

High Compute Cost

Fine-tuning and retraining require GPUs, infrastructure, and repeated experiments before a useful model is ready.

Slow Iteration

Every new task, customer, dataset, or domain requires another training cycle — slowing down product development and deployment.

Specialized Expertise

Building reliable custom models requires deep ML expertise, careful evaluation, and non-trivial engineering.

OUR SOLUTION

Pelora turns model adaptation into model generation.

Instead of training a new model for every task, Pelora learns how models should adapt. Our system creates compact model updates that can be applied directly to any base model, such that, training becomes dramatically cheaper.

01

Input examples + task

Pelora receives a small task-specific dataset, examples, constraints, or a natural-language task description.

02

Generate model update

The system generates task-specific adaptation weights instead of running a conventional training loop.

03

Deploy instantly

The generated update is applied to a base model and can be evaluated, deployed, or iterated on rapidly.

TECHNOLOGY

Learning the space of model updates.

Pelora is based on a new architecture for generating compact model updates. The system learns from trained adapters and their corresponding tasks, then uses that knowledge to produce new task-specific updates on demand.

Task & Data Conditioning

The system uses both semantically encoded examples and task descriptions to understand what adaptation is needed.

Generative Model Updates

Instead of optimizing weights from scratch, Pelora generates structured updates that adapt a base model to a new task.

Performance Feedback

Generated updates can be evaluated on real tasks, allowing the system to improve toward practical model performance.

USE CASES

For teams that need many specialized models.

Healthcare & Life Sciences

Generate models adapted to clinical notes, biomedical literature, genomic data, or specialized healthcare workflows.

Enterprise AI

Create models specialized for internal documents, customer support, compliance, legal workflows, or proprietary knowledge.

Research & AI Labs

Accelerate experimentation by replacing repeated fine-tuning runs with fast generated model updates.

WHY US

Built by AI researchers and ML systems builders.

Pelora is founded by Ben Shapira and Roi Cohen. Our work spans language models, model reliability, biomedical AI, adversarial robustness, and production ML systems. We combine frontier AI research with practical experience building and evaluating real-world AI systems.

0+ Publications

0 Top-Tier Venues

0 Industrial Research Labs

0+ Years Combined Experience

Selected Research

NeurIPS 2024 I Don't Know: Explicit Modeling of Uncertainty with an [IDK] Token Roi Cohen, Konstantin Dobler, Eden Biran, Gerard de Melo
TACL 2024 Evaluating the Ripple Effects of Knowledge Editing in Language Models Roi Cohen, Eden Biran, Ori Yoran, Amir Globerson, Mor Geva
EMNLP 2023 LM vs LM: Detecting Factual Errors via Cross-Examination Roi Cohen, May Hamri, Mor Geva, Amir Globerson
Findings EMNLP 2025 InFact: Informativeness Alignment for Improved LLM Factuality Roi Cohen, Russa Biswas, Gerard de Melo
arXiv 2024 MAMMAL: Molecular Aligned Multi-Modal Architecture and Language Yoel Shoshan et al., including Ben Shapira
arXiv 2025 BioVERSE: Representation Alignment of Biomedical Modalities to LLMs Ching-Huei Tsou et al., including Ben Shapira
Under Review StyleAT: Defending Face Recognition Against Semantic Attacks Ben Shapira, Roi Cohen, Shang-Tse Chen, Mahmood Sharif
Under Review Knowledgeless Language Models: Decoupling Linguistic Competence and Factual Knowledge Roi Cohen, Yvan Carre, Nick Lechtenborger et al.

THE TEAM

Meet the Founders

Ben Shapira

Co-Founder · AI Researcher

M.Sc. in Computer Science from Tel Aviv University. Research Scientist at IBM Research working on biomedical and multimodal AI. Previously built production ML, NLP, and LLM systems at AlgoSec. Former intelligence analyst at IDF Unit 8200.

Roi Cohen