Where Noise Ends,
Alpha Begins.
AlphaEnd Labs applies large language models to quantitative factor discovery — searching wide, selecting hard, and keeping only the signal that survives out of sample.
We mine the signal that survives the noise.
AlphaEnd Labs is a ten-person research lab bringing large language models to quantitative factor discovery. Our systems generate, mutate, and cross over factor hypotheses under structured quality gates — then distil them into a compact library of low-correlation alphas, evaluated with the discipline of a trading desk.
We believe the edge comes from search breadth times selection discipline — not from any single factor being clever.
What we build.
alpha-loop
An LLM-driven, self-evolving factor-mining framework — diversified parallel search, trajectory-level evolution, and structured quality gates, distilled into a validated factor library.
View repository →stock-benchmark
A benchmark that pits time-series architectures against one another as alpha combiners — turning a factor library into a single prediction, then a cost-aware portfolio.
View repository →On the roadmap — neutralization, portfolio construction, high-frequency signals, and end-to-end execution.
Method, not hype.
Search breadth
Every idea fans out into ten diversified planning directions, evolved through mutation and crossover — hundreds of candidates, not a handful.
Selection discipline
A multi-stage validated-pool pipeline — executable filtering, correlation de-duplication, adaptive thresholds — keeps only robust, low-correlation factors.
Evaluation hygiene
Mining feedback and final evaluation are strictly separated. The test years are never seen during search — so the numbers mean what they say.