Research assistant
Plan and execute research tasks. Deliver results, executable methods, and traceable records that scientists can review and refine.
PRODUCT ENTRY POINTFROM ASSISTANCE TO DISCOVERY
Research
that learns.
Life-science AI that develops better methods, learns from experimental feedback, and carries validated experience into the next discovery.
A research platform built around recursive self-improvement.
THE EVOFOUNDRY ENGINE01—04
Every result.
A better starting point.
A better starting point.
- 01Define the objectiveQuestion, data, constraints
- 02Develop the methodRoute, search, improve
- 03Execute & validateCompute meets experiment
- 04Retain what worksVerified methods, reusable experience
EXPERIMENTAL FEEDBACK / NEXT ITERATION
Architecture overview. The integrated RSI loop is under development.
22
Drug-property benchmark tasks
2
Applications with wet-lab validation
6,000+
Company-owned experimental trajectories
Company-reported research evidence · September 2026
01 / THE PLATFORM
A model and a platform.
Built to evolve together.
Research assistance brings in real tasks. High-quality computational and experimental trajectories inform model post-training. Validation connects those capabilities to discovery.
Life-science models
Use task trajectories and experimental feedback to advance reasoning, planning, and generalization through post-training and independent evaluation.
CAPABILITY DEVELOPMENTExperimental feedback
Connect candidate design with partner-lab measurements. Retain validated experience and selectively advance research pipelines in areas of strength.
REAL-WORLD VALIDATION02 / RECURSIVE SELF-IMPROVEMENT
Improve how research
gets done.
EvoFoundry’s research engine connects three layers of improvement: the experience it retains, the methods it develops, and the decisions it makes during execution.
Remember verified experience
Method memory, improvement memory, and execution memory preserve effective approaches, failed attempts, and repair paths.
Develop better methods
Combine scientific models with task-specific algorithms. Search for improvements, then train and validate them on each new task.
Adapt goals and execution
Update objectives with new evidence, route tasks to proven starting points, and coordinate tools, compute, evaluation, and recovery.
Research evidence supports individual modules. Integration of the complete RSI loop and periodic model post-training remain ongoing development work.
03 / EARLY VALIDATION
From computational results
to measured outcomes.
Two experimental applications provide an early foundation for a broader life-science platform. Computational benchmarks separately test method development and reuse.
Design. Screen.
Measure. Iterate.
17 / 107Experimental hits / candidates
3 weeksTwo screening rounds
Best candidate C3MRSA MIC 1 μg/mL
Study context
Company-reported early antibacterial screening. Hits used MIC ≤32 μg/mL. These are laboratory results; the reported MIC is specific to the tested candidate and assay.
Structure-informed design.
Measured binding.
3 / 24Candidates with PD-L1 binding
Apparent binding affinityKD 300–395 nM
Study context
Company-reported BLI measurements of 24 candidates identified three PD-L1 binders. The image is a structural prediction; binding evidence comes from the experimental assay.
Drug-property prediction
Aggregate normalized performance across 22 tasks, compared with nine agent approaches.
Lower LLM API cost
On six held-out computational tasks versus Claude Code. This comparison concerns API costs.
Memory combinations above baseline
Of tested experience subsets outperformed the no-memory baseline when the pool contained 13 tasks.
All figures are company-reported research results from September 2026. Computational benchmarks and wet-lab measurements are presented separately; results depend on the stated task and assay conditions.
04 / COMPOUNDING RESEARCH ASSETS
Keep the experience
behind the result.
Experimental facts, executable methods, and improvement records create reusable starting points. New tasks test whether that experience still adds value.
6,000+Company-owned wet-lab trajectories
1,000+Proprietary small- and large-molecule assets
Trajectory use and model iteration are scoped to authorized data.
05 / TEAM CAPABILITIES
Across models, agents,
and experiments.
EvoFoundry brings together model-training expertise, scientific agent development, and experimental collaboration to connect computational research with physical validation.
Model training
Foundation-model training, post-training, and evaluation across scientific tasks.
Scientific agents
Task planning, method search, tool use, and traceable execution.
Experimental collaboration
Candidate design, partner-lab validation, and structured experimental feedback.
06 / THE NEXT STAGE
Build the model.
Deepen the evidence.
Capital supports model post-training, core hiring, and computational–experimental validation as EvoFoundry advances its platform and selected research pipelines.
CUMULATIVE FINANCING TARGET$8–10M
Company plan · September 2026
First-generation product & assets
Research assistant and RSI engine, proprietary experimental trajectories and molecular assets, and two early wet-lab applications.
Data & post-training
Collect task trajectories, link experimental outcomes and failures, build independent evaluation sets, and complete a first post-training cycle.
Partnerships & feedback
Deepen academic and industry collaboration, run multiple validation rounds, and establish repeatable experimental feedback processes.
Pipelines & model iteration
Advance selected candidates, repeat and optimize experiments, develop patent opportunities, and evaluate cross-task model improvement.
Forward-looking development milestones are plans, rather than completed outcomes.
INVESTOR INQUIRIES