←←All cases

Gen AI chatbot for REDI: simplifying access to cancer information with Claude 3.5

Client

REDI—Ratner Early Detection Initiative

Industry

Healthcare and Medical Research

Services

Gen AI development, data preparation and integration, third-party AI platform customization, prompt engineering, Gen AI testing

Tech stack

Chatling.io, Claude API, Claude 3.5, OCR

Challenge

Early cancer detection can improve survival rates, expand treatment options, and reduce the cost of care. Yet people may postpone screening or seeking professional help because they lack awareness, cannot locate relevant information, or are hesitant to discuss their concerns with a healthcare provider. REDI, the Ratner Early Detection Initiative, is a nonprofit organization advancing cancer research and public education about early detection. Its website offered an extensive collection of educational resources, but visitors struggled to navigate the available material and find information relevant to their questions. REDI wanted to make this knowledge easier to access through a conversational interface. The organization approached ITRex to implement an AI-powered chatbot that could answer questions about cancer screenings, symptoms, risk factors, and treatment options in an accessible yet professional tone. Because the chatbot would provide public health information, it required clear safety boundaries. It could provide general educational advice but not diagnose conditions, recommend treatments, request private information, or act as a substitute for a healthcare professional. Speed was another major constraint: REDI wanted the chatbot developed, reviewed, and launched within three weeks.

ITRex was tasked with:

●

Preparing REDI’s educational materials for use as the chatbot’s knowledge base

●

Selecting and configuring an AI model capable of producing clear, concise, and natural responses

●

Creating an accessible editorial tone inspired by New York magazine

●

Adapting responses to the user’s apparent level of subject-matter knowledge

●

Displaying supporting sources and their authors to make answers more transparent

●

Establishing safeguards against medical advice and requests for private information

●

Displaying a non-medical advisory disclaimer when conversations involved symptoms or potential diagnoses

●

Allowing users to save conversations as links or share them with others

●

Customizing the interface around REDI’s visual identity and integrating it into the organization’s website

Solution

ITRex delivered the bespoke Gen AI solution in a single three-week sprint, with ongoing client feedback. Instead of building the chatbot infrastructure from the ground up, the team decided to use Chatling.io, a third-party no-code platform that possessed the necessary knowledge base, model, interface, and integration capabilities to meet the tight project deadline. Using Chatling.io, ITRex deployed Claude 3.5 to power the chatbot. The team configured the model, crafted its prompts, tested its responses, tailored the interface to REDI’s brand, and embedded the chatbot in the organization’s website. Here's how the project unfolded.

Week 1: Knowledge base preparation & model selection

ITRex created the chatbot’s knowledge base from REDI’s educational materials, which included DOCX files, PDFs, and web pages. We used optical character recognition where necessary to extract content from source documents and make it available to the chatbot. The team reviewed the project requirements and chose Claude 3.5 for its strong summarization capabilities and conversational writing style. The model's temperature was set to 0 to produce more deterministic and predictable results.

Week 2: Prompt engineering & safety controls

ITRex used prompt engineering to establish the chatbot's identity, mission, and tone. The prompts instructed Claude 3.5 to use the conversation to assess the user's familiarity with the topic and adjust the terminology and level of detail as needed. The team also established safeguards appropriate for a public cancer-information service:

●

General information only. The chatbot can explain cancer screenings, symptoms, risk factors, and treatment options, but it cannot make diagnoses or offer personalized medical advice.

●

Privacy-conscious conversations. We taught the chatbot not to ask for explicit private information from users.

●

Visible source attribution. Answers identify the relevant sources and authors so users can understand where the information comes from and explore the original material.

●

Non-medical advisory disclaimers. When users ask about symptoms or possible diagnoses, the chatbot shows a disclaimer stating that its response is informational and does not replace professional medical advice.

●

Language matched to the reader. The chatbot explains information in plain language to general visitors and provides more detailed answers, using appropriate medical terminology, to users who are already familiar with the subject.

In line with the client's HIPAA requirements, we instructed the chatbot not to request private information and to stay within clearly defined medical boundaries. We only launched it after testing confirmed that it followed these privacy and safety rules consistently.

Week 3: Testing, customization & deployment

We tested the Gen AI chatbot by asking it questions that REDI visitors would likely have. We also confirmed that its answers were relevant, easy to understand, properly sourced, and stayed within the agreed-upon medical information boundaries. REDI then reviewed the answers before approving the chatbot. Next, our team matched the interface to REDI's website using the organization's logo, colors, and visual style. We also set up the option to save conversations as links or share them with other users. The chatbot was added to the REDI website as a separate page. Visitors can now engage in natural conversation with it, asking cancer-related questions and learning more about the organization's educational resources.

Impact

In just three weeks, ITRex transformed REDI’s existing educational library into an interactive cancer-information resource powered by Claude 3.5.

●

More value from REDI’s existing content. The organization can now present the same research in a format that is easier to explore, whether a visitor has a general question or wants a more detailed explanation.

●

Greater awareness of early detection. By making information about screenings, symptoms, and risk factors easier to find and grasp, the chatbot supports REDI’s mission to encourage earlier conversations about cancer.

●

A clearer case for the benefits of early diagnosis. One study estimated that diagnosing stage III and IV cases across 19 cancers at stages I or II could reduce annual US cancer treatment costs by approximately $26 billion, or 17% of estimated spending. By making research like this easier to access and understand, the chatbot supports REDI’s mission to raise awareness of the importance of early cancer detection.

●

A trustworthy way to communicate sensitive information. Visitors can verify what they read by referring to original sources and authors, while disclaimers and clear medical boundaries separate educational information from diagnosis or medical advice.

A practical solution for a small nonprofit. By adapting an existing platform rather than building a new chatbot infrastructure, ITRex helped REDI launch a custom Gen AI solution quickly without an in-house technology team.