Building AI research agents for quantitative discovery.
I am Pengyi, a quant research engineer and AI builder focused on turning messy financial research workflows into reproducible systems: hypothesis generation, implementation, backtesting, diagnosis, and next-round research planning.
From real financial constraints to research automation.
My core direction is to build a public, reproducible research stack for AI-assisted quantitative discovery. The long-term goal is to become an AI scientist who can produce strong open-source systems, technical reports, and top-conference research.
Research Thesis
Quant research is a loop, not a one-off model. The valuable system is an agentic workflow that can repeatedly propose, implement, test, diagnose, and revise hypotheses.
- Factor hypothesis generation
- Backtest and bias diagnosis
- Research memory and experiment lineage
- Human PM review before production decisions
Career Thesis
The next position should strengthen both research credibility and engineering output. RA, research engineer, quant researcher, and AI-for-finance roles are all evaluated by this bar.
- Can it produce papers, reports, or public systems?
- Can it create strong recommendation signals?
- Can it preserve cash flow and deep-work time?
- Can it compound into PhD and open-source impact?
Current public-facing project portfolio.
These projects form one system: a personal research operating system for quant, AI agents, paper production, and application materials.
Pengyi Quant R&D Agent
Agent workflow for quant research: factor hypothesis, implementation, backtesting, bias diagnosis, and next research plan generation.
Pengyi Quant Research OS v0
A reproducible research operating system for experiments, reports, artifacts, and factor research lineage.
Auto Paper NorthPolestar
Paper-production workspace for turning project ideas, experiments, and benchmark results into technical reports and preprint-ready drafts.
Auto CV Space
Versioned CV and application package system for RA, PhD, AI research, and quant roles.
FICC AI Infra Demo
Public-safe demo direction for financial research automation, focused on architecture and workflow rather than confidential data.
Everything learned should become an asset.
This website is not only a CV page. It is a public index of learning, research output, engineering progress, and career compounding.
Research
Reading papers in LLM agents, AI for finance, time-series modelling, backtesting, and research automation.
Engineering
Building reproducible Python systems, CLI tools, experiment artifacts, documentation, tests, and public demos.
Writing
Converting projects into technical notes, benchmark reports, RA emails, PhD statements, and future preprints.
See the structured log: learning.html. Read the technical blog archive: /archives/.
Selected achievements and proof points.
WorldQuant IQC UK Rank
Quant alpha research signal and competitive proof point.
Mathematics & Data Science
Mathematical foundation for quantitative finance, statistics, and machine learning.
Research Infrastructure
Building an integrated project system across agents, experiments, CVs, notes, and paper drafts.
Current transition plan.
RA / research role pipeline
Apply to aligned labs and teams in AI agents, AI for finance, quant research automation, RAG, and research engineering.
Technical report and public demo
Package Quant R&D Agent and Research OS into a public-safe demo, with examples, tests, and a short technical report.
Use RA output to strengthen PhD applications
Target stronger recommendation letters, clearer research fit, and tangible project output before the next application cycle.
Open-source AI scientist system
Compound toward a public research system with benchmark results, papers, and an independent open-source identity.
Blog system and earlier website versions.
The blog is the long-term research ledger. Earlier website drafts are preserved as design and narrative experiments.
Technical Blog
Jekyll/Chirpy post archive for project notes, paper reading, RA/PhD planning, and AI scientist output.
Version A
Academic / PhD / research-first presentation.
Version B
Industry / quant engineer / builder-first presentation.
Version C
Universal researcher-builder presentation.