Chris Carlton
Some things about me:
- Grew up in Minnesota, 4th-generation Golden Gopher, spent the past 3 years in FL
- Worked in investment banking (mergers & acquisitions) during school, after grad I spent some time consulting for small businesses, most recently worked in investment research at a small family office
- Left March 2026; currently looking for my next role!
- Still interested in finance (of course) but I've also been having fun building with AI lately
- I will try to publish more fun stuff here while I search for my next role
What do I want?
- I love finance, data, and applied AI. I am happiest building things and analyzing data
- Prefer tight-knit teams, curious people, and meaningful problems
- Open to exploring a diverse opportunity set, so please don't hesitate to reach out!
Some side projects:
- AutoActuary an autonomous research harness for actuarial modeling based on Andrej Karpathy's 'AutoResearch' repository
- Mixed results: the Autonomous research agent achieved a 19.1% improvement in Gini coefficient compared to the baseline Generalized Linear Model (but suffered degradation in other metrics)
- 24-hours of autonomous research agent runtime, 200+ autonomous experiments ran
- Python-based autonomous agent which constantly hypothesizes and executes P&C pricing model improvement experiments in R using GLM & GBM (LightGBM) models
- ChatWithASOP allows actuaries to interactively chat with the Actuarial Standards of Practice, published by the Actuarial Standards Board
- Stores ASOP metadata, parsed PDF chunks, full-text indexes, and vector embeddings in Supabase PostgresQL vector database
- Searches with a hybrid agentic search function using GPT-5-mini that combines full-text ranking and semantic search
- Routes common ASOP phrasing, glossary terms, and specific ASOP references into more targeted search plans
- Returns source citations, excerpts, and a bounded search trace alongside each answer
- IB-PPTX-Skill distills the best practices from investment banking PowerPoint deck creation workflows into a reusable skill for Claude/Codex so that every deck you create communicates with 'board room' level efficacy
- Clean, professional, 'board ready' deck template
- Slides lead with one main takeaway in the title
- Professional fonts, consisent font sizing, symmetrical object alignment
- GBM-Skill builds Gradient Boosting Machines for property and casualty pricing that work out of the box!
- 'Expert Mode' for sophisticated users/actuaries, agentic 'Beginner Mode' for new users/actuaries who aren't sure which model inputs to use
- Give Codex the model inputs and assumptions you want to use (Expert Mode) or let Codex reason over your data sources to extract which model assumptions will best achieve your target goal (Beginner Mode)
- Creates Python-based modeling workflows for frequency, severity, loss cost, and pure premium work, including exposure treatment, diagnostics, relativities, documentation, and optional Excel exhibits.
- 'Expert Mode' for sophisticated users/actuaries, agentic 'Beginner Mode' for new users/actuaries who aren't sure which model inputs to use
- GLM-Skill builds Generalized Linear Models for property and casualty pricing that work out of the box!
- Give Codex the model inputs and assumptions you want to use (Expert Mode) or let Codex reason over your data sources to extract which model assumptions will best achieve your target goal (Beginner Mode)
- Give Codex the model inputs and assumptions you want to use (Expert Mode) or let Codex reason over your data sources to extract which model assumptions will best achieve your target goal (Beginner Mode)
- Creates R-based modeling workflows for frequency, severity, loss cost, and pure premium work, including exposure treatment, diagnostics, relativities, documentation, and optional Excel exhibits.
- Give Codex the model inputs and assumptions you want to use (Expert Mode) or let Codex reason over your data sources to extract which model assumptions will best achieve your target goal (Beginner Mode)
- WordPuzzles an unlimited, shareable version of Wordle!
- Recreation of Wordle (Josh Wardle/NYT) and WordPlay (Dharmesh Shah) - my first working codebase/fun project assisted by Claude 3.5 Sonnet
- Unlimited gameplay, no need to wait for the next day's Wordle puzzle
- Game ID sharing - send a wordle puzzle to a friend (or friends!) to solve
- Statistics tracking for users who make an account so you can keep track of your performance over time
More links: