Skip to work
portfolio.

open to AI product, strategy & research roles

AI × Finance

research. build. decide.

Finance-trained builder exploring how AI agents can improve research, analysis and high-stakes decisions. Two live products, one LLM evaluation framework, and the research habits behind them.

view my work →let's talk
~/evidence — ship-log

open living-thesis

agentic research workflow · live

open cc-macro

falsifiable macro monitor · live

eval llms --domain finance

7-step benchmark · sanitized case

  • agent workflows
  • LLM evaluation
  • equity research
  • unit economics
  • human-in-the-loop
  • data lineage
  • falsifiable theses
  • PE diligence

// 01 — work

selected work.

Two products running in production and one evaluation framework. Each answers a different question about putting AI into research and decision-making.

~/products/living-thesis01

Agentic research & decision workflow

Living Investment Thesis

An investment thesis goes stale the day it is written. This keeps it alive: new evidence is routed to exactly the assumptions it touches, and the analyst decides what changes.

  • Thesis split into key drivers, falsifiers and a “what to watch” list
  • Relevance routing: only the affected research is re-evaluated
  • “What Changed” view: accept, edit, reject or leave unresolved, with version history

How a long-lived agent absorbs new information, when it should re-think, and where the human keeps the decision.

  • agent loop
  • human-in-the-loop
  • versioning
  • eval
live demo (opens in a new tab)
~/products/cc-macro02

Auditable, falsifiable macro research system

CC Macro Analysis OS

Is China forming a sustainable turning point in nominal growth? A macro call built so it can be audited, refreshed and proven wrong.

  • Preregistered thesis with competing hypotheses, stated before the data
  • Every claim bound to its data, with strict as-of discipline
  • Deterministic E0–E5 evaluation and append-only updates

Treats hallucination, data lineage, evaluation and the automation boundary as product requirements.

  • falsifiability
  • data lineage
  • deterministic eval
  • EN / 中文
live monitor (opens in a new tab)
~/cases/llm-eval-finance03

Enterprise AI capability benchmark

Evaluating LLMs for finance

How do you tell whether a language model is reliable enough for real financial workflows, and what should an institution actually deploy?

  • Task taxonomy and gold standards drawn from real finance work
  • One rubric plus a fatal-failure taxonomy across models
  • Security and data boundaries, then a deployment recommendation

Model evaluation and enterprise adoption judgment: the part of AI product work that decides what ships.

  • evaluation
  • rubrics
  • failure modes
  • enterprise AI
sanitized context & outcome

// 01b — research

the business judgment behind the builds

~/research/temu04

Business model & unit economics

Temu

Taking apart a new, complex business model: unit economics, competitive dynamics and what has to be true for it to work.

  • unit economics
  • competition
  • business model
report file or summary
~/research/pet-food05

PE-style industry & company research

Pet food

Investor-grade diligence on a consumer category, from industry map and expert interviews to forecast, valuation, returns and downside.

  • industry map
  • interviews
  • valuation
  • IRR / MOIC
report file or summary

// 02 — background

finance first, now building.

  1. feat: ship AI-native research systems

    Building AI × Finance products

    organisationdates

    • Designed and shipped Living Investment Thesis, an agentic research and decision workflow
    • Designed and shipped CC Macro Analysis OS, a falsifiable, auditable macro research system
    • Built an LLM evaluation framework for high-stakes financial tasks
    • agents
    • evaluation
    • product design
  2. commit message from résumé
    role title

    organisationdates

    3–5 bullets from résumé

  3. commit message from résumé
    role title

    organisationdates

    3–5 bullets from résumé

HEAD → building AI-native research tools

résumé (EN) PDF简历（中文）PDF

// 03 — skills

what I work with.

— ai product05

  • agent loop design
  • evaluation rubrics
  • human-in-the-loop UX
  • failure taxonomies
  • AI-assisted building

— finance05

  • DCF valuation
  • LBO modelling
  • real-estate PE models
  • unit economics
  • IRR / MOIC

— research04

  • industry mapping
  • expert interviews
  • falsifiable theses
  • macro data

— tools00

list from résumé

// 04 — contact

let's talk.

Open to AI product, AI strategy and investment research roles. Email is the fastest way to reach me.

public email address
  • LinkedIn URL
  • résumé (EN)
  • 简历（中文）

open to new roles

hello.eml
to:
email
subject:
Portfolio inquiry

Hi, I looked through your work and would like to talk about a role on our team.

needs email