Company Performance Metrics
- Wasif Masood: Founder
ChaperoneAI builds enterprise-grade, domain-specific large language models for the pharmaceutical and biotech industry. Its custom-trained LLMs automate the extraction, summarization, and classification of complex scientific content — clinical trial data, patents, regulatory documents, and publications — replacing manual processes that are slow,
costly, and error-prone.
The company has developed and publicly released multiple LLMs on Hugging Face, surpassing 27,000 downloads. Its flagship model, Chaperone-Thinking-LQ-1.0, is a 32B-parameter reasoning model compressed to ~20GB via a four-stage GPTQ-QAT and QLoRA pipeline — achieving 84% on MedQA (within 4 points of GPT-4o), 91.9% on MATH-500, and 85.9% on MMLU, while running on a single NVIDIA L40/A100 GPU with 1.6x higher throughput and 43% lower latency than its full-precision base model. Fully open-source under CC-BY-4.0.
ChaperoneAI also powers Deep Research (deepresearch.chaperoneai.net), an AI-powered research copilot that searches 200M+ academic papers across OpenAlex, arXiv, and PubMed, synthesizing findings into citation-backed reports — running entirely on ChaperoneAI's self-hosted inference infrastructure with no third-party LLM dependencies. Revenue is generated through custom model fine-tuning engagements, subscription-based API access, and on-premise licensing for clients with strict data governance requirements (GDPR/HIPAA). Clients include R&D, medical affairs, and commercial operations teams across pharma.
ChaperoneAI operates under Empirisch Tech GmbH (Vienna, Austria) and is backed by Microsoft for Startups (Stage 5, $150K Azure credits), NVIDIA Inception, and AWS Startup programs.
Website: chaperoneai.net
Models: https://huggingface.co/models
