Biotech Orchestration System
Reflecting the multi-billion dollar deal between Insilico Medicine and Eli Lilly, this challenge involves building a high-performance biotech agent orchestrator. You will use the Google ADK to develop a system that coordinates drug discovery tasks across various specialized environments. The core reasoning engine will be Gemini 3.1 Pro, which will use its large context window and multimodal capabilities to analyze molecular structures and research papers. You will integrate Featherless AI for high-throughput model serving of specialized chemistry models and Edge Impulse to handle data from lab sensors. This system will simulate the orchestration of a co-development pipeline, where agents must determine the feasibility of drug candidates based on simulated efficacy data. The challenge emphasizes the use of modern model runtimes and specialized inference for high-stakes scientific applications.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a multimodal biotech drug discovery agent using Google ADK and Gemini 3.1 Pro for molecular analysis and simulation orchestration.
How work is evaluated
Evaluates the agent's accuracy in drug candidate screening and simulation management.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
Runtime Routing Test
Ensure requests are correctly routed to Featherless AI endpoints
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Analysis Depth
Measures the detail and scientific accuracy of the reasoning • target: 9 • range: 0-10
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Deploy Gemini 3.1 Pro with extended reasoning for analyzing complex biochemical pathways
Master Google ADK for building agents that leverage Vertex AI and multimodal tool-calling
Utilize Featherless AI to run and route requests to fine-tuned Llama 4 Maverick models for chemistry tasks
Integrate Edge Impulse to ingest and process real-time sensor data from drug trials within the agent loop
Design hierarchical planning workflows where a lead agent delegates molecular docking simulations to specialized worker agents
Implement Claude Sonnet 4.6.6 as a cross-verification agent to peer-review Gemini 3.1 Pro research findings
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
Evaluates the agent's accuracy in drug candidate screening and simulation management.
Challenge input
SMILES string and molecular image
Google ADK
Agent Development Kit for
Google Gemini
Google's multimodal AI model
Google AI model provider
Evaluated output
Feasibility score and reasoning
- Ensure requests are correctly routed to Featherless AI endpoints
- Measures the detail and scientific accuracy of the reasoning • target: 9 • range: 0-10
- Analysis Depth target: 9
- 1 public reference case
- Docker execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Google ADK · Required
- Google Gemini · Optional
- Google · Optional
Evaluation contract
- Runtime Routing Test · Weight 1
- Analysis Depth · Weight 1
Metrics
- Analysis Depth · Target 9
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: SMILES string and molecular image
Output: Feasibility score and reasoning
Execution harness: Docker
Recipe freeze marker
No recipe freeze time is recorded.