← back
Morgan Stanley's ALPHALAB: Multi-Agent Research Across Optimization Domains — Brendan Rappazzo
Takeaway
Automated research needs a harness that connects hypotheses, trustworthy evaluations, compute access, and reusable domain context.
Summary
- Morgan Stanley's ALPHALAB automates quant research from a dataset and prediction objective through research, evaluation construction, and large-scale experimentation.
- The custom harness is provider-agnostic and designed to integrate with existing data, backtesting, and production models.
- Core tools include shell access, web search, and Slurm job submission for GPU experiments.
- Research to-do lists and Markdown notes establish reusable context and let the system keep working when an agent would otherwise stop early.
alphalabautomated-researchquantitative-finance
Original description
When coding agents got good enough at the end of 2025, Morgan Stanley's roughly thirty person research group asked what would happen if agents ran the research, not just wrote the code. The result is AlphaLab, a multi agent system they built and open sourced. You hand it a problem in plain language and it writes the code, sets up back tests and evals, configures and submits cluster jobs, and runs the statistical tests, managing its own context as it goes. They skipped every off the shelf framework and wrote their own harness so they could watch how it reasons and bake in their own standards. The shape is a strategist that proposes experiments and workers that run them, laid out as a board of cards people can read, edit, and approve before the loop optimizes against a weak eval. Rappazzo shows it finding real gains now used internally, from fine tuning a model to predicting credit bonds, and argues the lasting human job is designing the verifiable environment the agents compete in, like a private Kaggle, while they handle the middle. Speaker info: https://x.com/brendanh0gan / brendan-rappazzo-hogan-763734115 https://www.bhogan.net Timestamps: 0:00 - Introduction: a thirty person research group 1:30 - What changed when coding agents arrived 2:55 - Building AlphaLab 1.0, open sourced 4:34 - Encoding enterprise standards into the system 6:40 - Why they skipped off the shelf harnesses 8:21 - How the research loop runs 9:36 - Strategist and worker agents 10:40 - Guarding against a bad eval 13:23 - Finding real improvements 16:12 - Verifiable environments, like Kaggle 18:45 - The human job in the limit