Building Better Cancer Treatment Plans with Dynamiq’s Multi-Agent AI

Summary

A major healthcare provider sought a way to support surgeons and clinicians in creating more informed, personalized treatment plans for cancer patients. They partnered with Dynamiq to build a multi-agent AI assistant that combines LLMs with real-time access to clinical research, guidelines, and trial data.

Built entirely with Dynamiq’s low-code agentic AI platform, the solution integrates voice input, clinical guidelines, and PubMed search — all powered under the hood by Anthropic Claude 3.7 Sonnet via AWS Bedrock, Amazon Titan Text Embeddings v2, and an Amazon OpenSearch as a Vector Database — to deliver evidence-based treatment plans that beat standard LLMs in accuracy and reliability.

The Challenge

Clinical decision-making in oncology requires fast access to constantly evolving information: the latest NCCN and ASCO guidelines, stage 3 clinical trials, and peer-reviewed research. But surgeons and clinicians often lack the time and bandwidth to search and learn all of this information.

That is why our client needed an AI solution that could:

• Pull data in real time from trusted medical sources like PubMed
• Curate cancer treatment plans using the most up-to-date research
• Understand nuanced patient histories, including prior treatments and comorbidities
• Support reasoning and multi-step decision-making
• Integrate voice-to-text for hands-free interaction during consultations

Rather than building a complex system from scratch, the client wanted a solution that would let them focus on clinical logic and data, not infrastructure issues.

Why Dynamiq

The healthcare provider chose Dynamiq after a trusted partner recommended the platform. Its ability to support the development and deployment of complex, multi-agent systems without requiring a large internal AI team or custom infrastructure was precisely what the client needed.

Dynamiq’s platform stood out because it offered:

• A visual low-code interface that made it easy to design multi-step agent workflows and connect them to medical data sources
• Out-of-the-box voice-to-text capabilities, ideal for hands-free use in clinical settings
• Built-in support for retrieval-augmented generation (RAG), allowing the assistant to combine live research data with language model reasoning
• Seamless integration with external sources like PubMed, NCCN, and ASCO, without custom pipelines
• Out-of-the-box connectors to AWS Bedrock services - Anthropic Claude 3.7 Sonnet, Amazon Titan Text Embeddings v2, and the Amazon OpenSearch as a Vector Database, eliminating custom DevOps work.

For the client’s team, it meant they could move fast, keep control of the implementation, and deliver a working solution that met clinical needs.

The Solution

Using Dynamiq’s GenAI platform, the client’s team built a clinician-facing AI assistant designed to support oncologists in making personalized treatment decisions. What makes this solution unique is its multi-agent architecture—a coordinated system where each agent performs a specialized role in the decision-making process. Each AI agent calls Anthropic Claude 3.7 Sonnet for reasoning and writes/reads embeddings generated by Amazon Titan Text Embeddings v2 from an Amazon OpenSearch Vector DB index, enabling low-latency RAG at scale.

The assistant operates in several stages:

Clarification Agent

The first agent gathers detailed information from the clinician about the patient’s condition, medical history, previous treatments, lab results, and even socio-economic factors.

Guideline Retrieval Agent

The next agent searches relevant oncology treatment guidelines, including the latest updates from NCCN and ASCO, based on the patient’s diagnosis and clinical profile.

Clinical Trial Agent

Next, an AI agent pulls up-to-date results from stage 3 clinical trials, focusing on emerging therapies and drugs relevant to the patient’s specific case.

Treatment Planning Agent

Finally, this AI agent synthesizes all the collected data (context, guidelines, research) into a detailed treatment proposal, complete with reasoning and links to supporting sources.

All agents are configured through Dynamiq’s low-code interface, making it easy to adjust workflows, refine logic, and add new capabilities without custom development. 

The entire system was implemented by one forward-deployed ML engineer from Dynamiq, working alongside the client’s internal team.

Results

• End-to-end cancer treatment AI assistant built and deployed using Dynamiq’s platform
• Outperformed vanilla LLMs in treatment recommendation accuracy
• Seamless voice input integration, optimized for clinician workflows
• Fully functional multi-agent reasoning pipeline configured in Dynamiq’s builder

What’s Next

The healthcare provider is now expanding its AI capabilities with Dynamiq, building additional AI agents to support clinicians across various specialties. The goal is to reduce cognitive load, increase accuracy, and deliver better patient outcomes.