Fearn

What Fearn does

Fearn is an AI-native patent prosecution firm founded by former Big Law patent experts and PhDs in artificial intelligence. It develops proprietary, hypercompact foundation models for intellectual property law, offering unmatched efficiency and data sovereignty. These models are engineered to run on virtually any hardware, enabling secure deployment without compromising performance. Fearn aims to reshape how the world creates, protects, and powers ideas by providing fast, reliable, and cost-effective patent drafting and prosecution services, from first draft through examination. The company emphasizes data security and compliance, using zero outside AI models to ensure data privacy and prevent public-disclosure issues.

Key capabilities

• Proprietary, hypercompact foundation models
• Unmatched efficiency and data sovereignty
• Models run on virtually any hardware
• Secure deployment without compromising performance
• Fast patent application process (approx. 3 days to file provisional)
• Expert drafting by former Big Law patent experts
• Flat-fee pricing with no surprise legal bills
• Guarantee on non-provisional patent costs if no claims are allowed
• Zero outside AI models for data privacy
• Real-time feedback during invention explanation
• Ability to handle complex, structured, and cross-referential patent applications
• Compositional approach based on neurosymbolic architectures
• Small, hand-corrected, hand-labeled dataset for model training

Use cases

• Drafting provisional patent applications
• Drafting non-provisional PCT patent applications
• Entering US national phase from PCT applications
• Responding to office actions from patent examiners
• Building patent portfolios for high-growth startups
• Accelerating patent filing timelines for biotech and hardware companies
• Securing intellectual property for new inventions
• Protecting innovations at pre-seed and seed stages for funding rounds

AI approach

Fearn develops proprietary, hypercompact foundation models engineered to run on virtually any hardware for secure deployment without compromising performance. They use a compositional approach based on neurosymbolic architectures, combining dozens of small LLM and non-LLM models, some built from scratch. This approach emphasizes control and resistance to hallucination. They also focus on small, hand-corrected, hand-labeled datasets for training, exploiting an inversion where models get more from carefully labeled examples than generic architectures get from noisy ones. They are pushing the frontier of VLMs for the patent lifecycle, working across language, image, and multimodal models, and using post-training techniques like RLVR.

Tech named: proprietary, hypercompact foundation models, compositional approach based on neurosymbolic architectures, small LLM and non-LLM models, hand-corrected, hand-labeled dataset, VLMs, RLVR, quantization, batching, distributed training, vLLM, SGLang, JAX, PyTorch

Industries served

• Software Development
• Robotics
• Biotech
• Hardware
• Pharmaceutical
• Life Sciences

What it says sets it apart

• AI-native firm with proprietary, hypercompact foundation models
• Founded by former Big Law patent experts and AI PhDs
• Significantly faster time to file patents compared to traditional firms
• Flat-fee pricing with no hourly billing and transparent costs
• Cost guarantee for non-provisional patents (return $9,000 if no claims allowed)
• High data security and privacy with zero outside AI models
• Compliance with ISO 27701, GDPR, SOC II, USDP, and ISO 27001 standards
• Focus on data sovereignty and secure deployment
• Models built from scratch using a compositional approach (neurosymbolic architectures) to reduce hallucination
• Leverages small, hand-corrected, hand-labeled datasets for high reliability
• Provides real-time feedback and interactive drafting process
• Ability to handle complex technical and legal nuances in patent drafting

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.