PERSONAL MASTER PRESENTATION / V0.1
Pengyi Peng
I build systems that make research ideas testable.
01 / IDENTITY
Research questions become stronger when they produce inspectable systems.
My work connects AI-native systems, quantitative research, and software engineering through evidence, tests, and deployment.
02 / JOURNEY
Mathematics gave me the language for state, uncertainty, and feedback.
Applied mathematics led to complex systems modelling, then to building AI and quantitative research infrastructure.
- 2018-2022 / Mathematics and Applied Mathematics
- 2022-2024 / MSc Complex Systems Modelling, King's College London
- 2025-now / Systems, research infrastructure, and open source
03 / METHOD
I organize unfamiliar systems through the same core questions.
Data -> Type -> State -> Invariant -> Transition -> Interface -> Boundary -> Architecture
- Model the domain
- Make failure visible
- Build the smallest working loop
- Measure and revise
04 / AI SYSTEMS
Agent capability needs routing, verification, and Human Gates.
I focus on tool use, memory, retrieval, planning harnesses, evaluation, and governed multi-model execution.
- Capability routing
- Evidence-aware memory
- Sandboxed execution
- Evaluation before release
05 / QUANT SYSTEMS
A research signal matters only when the whole path is explicit.
Data -> Feature -> Signal -> Portfolio -> Risk -> Execution -> Attribution
- Fixed income and factors
- Market microstructure
- Research-to-production boundaries
06 / INFRASTRUCTURE
The interface is part of the research system.
PWEB joins a typed frontend, Worker API, persistent data, testing, deployment, and evidence gates in one working artifact.
- React + TypeScript
- Hono + Cloudflare Workers
- D1 persistence
- Unit, integration, and E2E verification
07 / OPEN SOURCE
Small, bounded upstream changes are a real engineering test.
I use maintainer review to test code reading, boundary judgment, validation, and communication.
- LightRAG #3607 / token-limit boundary
- NautilusTrader #4637 / deterministic request tests
- Tencent WeKnora #2549 / archived-page statistics
08 / SELECTED SYSTEMS
The portfolio is a set of working research interfaces.
Each artifact exposes a different layer: full stack, fixed income, career evidence, and agent-native tooling.
- PWEB / full-stack learning and deployment
- Rates Bond Quant / fixed-income laboratory
- PENGYICV / canonical evidence packaging
- dsh-quant / bounded agent plugin
09 / RESEARCH AGENDA
The next questions sit between agent systems and research infrastructure.
How can agents retrieve the right capability, preserve evidence, and improve without bypassing evaluation?
- Outcome-aware capability retrieval
- Memory and provenance
- Harness and benchmark design
- Human-agent decision boundaries
10 / FIT
I am looking for an ecosystem where research and building reinforce each other.
The strongest fit combines ambitious questions, open technical exchange, rigorous evaluation, and systems that reach real users.
11 / NEXT CONVERSATION
Let us identify one question worth building and testing together.
Research collaboration / PhD conversation / AI and quantitative systems
- pengyi-peng.pages.dev
- github.com/pengpengyi92
- pengpengyi92@gmail.com