Solution
Roko Labs collaborated closely with the client subject matter experts to define project requirements and business logic. Through in-depth discovery sessions, the team identified key workflows and obtained source documents essential for training AI models and refining prompts.
Roko’s AI engineers designed a robust system architecture centered around a fine-tuned single model, incorporating Retrieval Augmented Generation (RAG). This approach enabled the system to effectively process domain-specific knowledge, ensuring precise extraction of key information from complex medical and regulatory documents. The system was developed and rigorously trained through iterative review cycles with the client’s experts, refining outputs until an optimal accuracy threshold was achieved. A custom-built UI was implemented to facilitate seamless document uploading, AI processing, and expert review—ensuring a streamlined workflow for generating high-quality pre-IND documents.
Roko Labs developed the Synthesis App for the client, a custom UI that enables the client to securely upload emails, text, and even video conversations for processing. These materials are stored using scalable solutions like AWS S3, with robust encryption to protect highly confidential data. During processing, the system first retrieves relevant data needed for generating a sub-section of an investigational new drug (IND) application by extracting key methods, procedures, and equipment details, including keywords, summaries, and structured data. Next, the system retrieves validation parameters such as accuracy, precision, and robustness, along with related testing data. Finally, the system compiles the retrieved information into a clear and structured document, ensuring a well-organized output that meets regulatory and industry standards.
Result
80%
accuracy score of generated documents from proof of concept focusing on two key IND sections.
50%
Reduction in time required per IND, creating a scalable framework for the client.
