Genesis

Genesis — Accelerating Cancer Research Through Simulation

Inspiration

Cancer research operates on timelines that don’t match the urgency of the disease itself. While tumors can progress in weeks or months, new treatments often take 12–15 years to reach patients due to the complexity of the research → trials → approval → clinical adoption pipeline.

This gap creates a critical bottleneck: researchers are forced to move slowly in a system that demands speed.

Genesis was inspired by a simple question: What if researchers could test and validate ideas at the speed of computation instead of clinical timelines?

We set out to combine digital twins, computational oncology, and real-time scientific evidence into a single system that reduces friction across the research workflow.

Problem

Modern cancer research faces three core challenges:

Fragmentation — Simulation, literature review, and hypothesis testing exist in separate tools
Latency — Testing even a single hypothesis can take months or years
Lack of iteration — Researchers cannot rapidly explore “what-if” scenarios at scale

As a result, promising ideas are slow to validate, and delays compound across the entire pipeline—ultimately affecting patient outcomes.

Solution

Genesis is an AI-powered digital twin platform designed to accelerate oncology research through simulation, evidence retrieval, and interactive experimentation.

It enables researchers to:

Model patient-specific disease progression
Simulate treatment strategies in real time
Retrieve and synthesize scientific literature
Compare experimental outcomes across scenarios
Build a persistent, iterative research workflow

By unifying these capabilities, Genesis transforms research from a linear process into a fast, exploratory system.

Core Capabilities

1. Patient-Aware Simulation

Genesis models disease progression using patient-specific parameters such as mutation status and tumor biology, rather than relying on population averages.

This allows researchers to explore how different interventions affect individual trajectories over time.

2. Evidence-Grounded Intelligence

Every output generated by Genesis is tied to real, cited biomedical literature.

The platform retrieves data from sources such as:

PubMed
Europe PMC
ClinicalTrials.gov
Semantic Scholar

This ensures transparency, traceability, and scientific credibility.

3. Interactive Research Workflow

Genesis enables rapid experimentation through an interactive interface where researchers can:

Apply and stack treatments (e.g., chemotherapy, immunotherapy, targeted therapies)
Simulate biomarker changes over 6–24 month periods
Compare baseline vs intervention scenarios
Iterate instantly without waiting for physical trials

4. Memory-Driven System

All simulations and experiments can be saved into a persistent notebook system.

This allows researchers to:

Track experimental history
Compare multiple treatment strategies
Build on prior insights over time

5. Explainable AI

Genesis provides structured explanations for every output, including:

Biomarker-level changes
Supporting evidence
Reasoning steps

This ensures users can understand not just what happened, but why.

System Architecture

Genesis is built as a multi-agent AI system designed for modular reasoning and scalability.

Key Components:

Supervisor Agent
Routes user requests to appropriate specialized agents
Digital Twin Agent
Simulates disease progression and treatment response
Research Agent
Retrieves and synthesizes biomedical literature
RAG Agent (Retrieval-Augmented Generation)
Queries vector databases (LanceDB) for relevant context
Review Agent
Validates outputs and cross-checks evidence

Flow:

User Input → Supervisor Routing → Specialized Agents → Validation → Structured Output

This architecture enables:

Separation of concerns
Scalable reasoning workflows
Integration of real-time data sources

Technology Stack

Frontend

Next.js (React framework)
TypeScript
Tailwind CSS
Recharts (data visualization)
Radix UI (accessible components)

Backend / AI Layer

Vercel AI SDK
Mistral (primary LLM for reasoning and synthesis)
Zod (schema validation for structured outputs)

Data & Retrieval

PubMed, Europe PMC, ClinicalTrials.gov
Semantic Scholar
LanceDB (vector database for retrieval)

Additional Systems

Local simulation fallback engine
Experiment store (JSON-based)
Notebook logging system

Multimodal & Multi-Provider Capabilities

Genesis supports:

Multimodal interaction
- Speech-to-text (user queries)
- Text-to-speech (AI responses)
Multi-provider AI models
Researchers can select models based on:
- Speed
- Cost
- Reasoning depth

This flexibility allows the system to adapt to different research needs and constraints.

Security & Safety

Genesis is designed with safety enforced at the system level.

Data Protection

Pre-flight PII detection blocks sensitive data (SSN, DOB, addresses)
No patient identifiers are sent to external APIs

Behavioral Guardrails

Blocks requests for MRNs, insurance IDs, and direct identifiers
Restricts off-topic or unsafe usage

System Safeguards

No bulk data export endpoints
Session-scoped data only
API key and credential leak detection

Validation Strategy

1. Simulation Validation

Benchmarked against published TNBC cohort data
Compared with known treatment outcomes

2. Evidence Validation

Only verified biomedical sources are used
Outputs are structured and schema-validated
Cross-checked by a review agent

3. System Reliability

Confidence scoring for outputs
Risk flags for uncertainty
Fallback to deterministic local simulation when needed

Challenges

Ensuring outputs felt scientifically grounded rather than purely generative
Coordinating multiple agents (simulation, research, validation) in real time
Maintaining performance while handling data retrieval and visualization
Designing an intuitive interface for a complex research workflow
Balancing system capability with strict safety constraints

Accomplishments

Built a fully functional multi-agent AI research platform
Enabled real-time simulation of TNBC progression
Integrated live biomedical literature retrieval
Created a unified workflow for simulation, validation, and iteration
Developed a persistent notebook system for tracking experiments

What We Learned

Structure and constraints (schemas, validation) are critical for reliable AI systems
Evidence grounding significantly increases trust and usability
Digital twins are far more powerful when combined with live data
Interactive systems enable faster and more meaningful exploration than static tools