Trading at the speed of… chat?

Background & Personal Experience

• Chloe Casale's Injury: Casale suffered a tibia stress fracture during the 10-day, 1,000km Mongol Derby (a long-distance horse race in Mongolia).
  • Cause: A stirrup banging against her shin caused a bruise; she stuffed socks into her pants for protection, which prevented ankle flexion necessary for riding.
  • Outcome: She finished the race on Day 10 with only 60km remaining, unable to put weight on her leg, but was allowed to ride because she could mount the horse independently.
• Race Statistics & Risks:
  • Completion Rate: Only 23 of 43 participants finished the race.
  • Primary Dropout Causes: Heat exhaustion (the hottest year on record), horses throwing riders (broken arms/legs), and kicks to the ground.
  • Casale's Path: She did not have a competitive riding history but was drawn to the remote navigation aspects of the event.

Career Trajectory

• Education Pivot: Initially intended to be a mechanical engineer (skipped high school chemistry for auto-tech), but pivoted to "Applied Computational Mathematic Sciences" at the University of Washington after winning a $20 bet regarding a Computer Science course.
• Palantir Tenure (4.5 years):
  • Role: Forward-Deployed Engineer (FDE/Delta).
  • Function: Embedded on-site with customers to solve specific problems, gather requirements, and feed insights back to product teams.
  • Scope: Worked across Brazil, Indonesia, Mexico, London, and Seattle; lived entirely out of one suitcase.
• Jane Street Transition:
  • Motivation: Joined after realizing limited growth at Palantir; impressed by Jane Street's rigorous hiring process (specifically the use of two interviewers per session).
  • Location: Moved to London, citing New York as "too soft" for her after experiencing the pace of the city.
  • Current Role: Desk Engineer for the Sales Trading team.

Sales Trading & Market Dynamics

• Sales Trader vs. Prop Trader Roles:
  • Sales Traders: Manage client relationships (e.g., pension funds, wealth managers), facilitate bilateral trades, and negotiate quotes via chat/phone to execute large orders.
  • Prop Traders: Focus on asset pricing, hedging, and building trading strategies; less client-facing, more focused on internal market making.
• The "Large Trade" Problem:
  • Adverse Selection Risk: Anonymous exchanges suffer from adverse selection where informed traders exploit market makers; this limits liquidity and widens spreads.
  • Bilateral Solution: Trading with known clients allows Jane Street to understand the reason for the trade (e.g., rebalancing), reducing adverse selection risk and allowing for better pricing than the open exchange.
• Chat vs. RFQ Platforms:
  • Client Preference: Many clients prefer direct chat (Bloomberg Terminal) over Request for Quote (RFQ) platforms to avoid integration costs and gain flexibility (e.g., requesting quotes without specifying size).
  • Efficiency: Automated parsing of chat messages reduces "fat finger" errors and speeds up quote response times (targeting <5 to <20 seconds for some counterparties).

Technology Stack & Engineering Practices

• Chat Parsing Evolution:
  • Legacy Approach: Used Regex parsers for standardized, high-frequency formats.
  • LLM Integration: Implementing Large Language Models to handle unstructured, typo-ridden, or non-standard chat formats.
  • Safety Risks: Concerns regarding "meeting of the minds" (legal contract formation) if an LLM misparses a trade request; potential for accidental trades at wrong prices or symbols.
  • Mitigation Strategies:
    • Ambiguity Reduction: Requiring traders to reply with non-ambiguous strings including security details, settlement dates, and explicit quantities.
    • Human-in-the-Loop: Using LLMs to assist but not replace human verification; testing parsers by inserting artificial errors to measure "alert fatigue."
    • Confidence Thresholds: Ensuring LLM accuracy approaches 99% before relying on automation, or using redundant prompts to verify parsing consistency.
• UI Development Frameworks:
  • Bonsai: An internal React-like framework built on OCaml for complex, production-grade user interfaces.
  • Webs: A distributed computational graph system (inspired by spreadsheets) for rapid data analysis and view creation.
    • Features: Allows non-engineers to build "views" and dashboards; supports collaboration and incremental computation.
    • Recruiting Integration: Used by recruiting to build views for candidate data; leverages AI code assistants for onboarding non-technical users.
  • Vibe Coding: Engineers and traders can "vibe code" (write ad-hoc HTML/JS) for quick prototypes, with a tiered scrutiny model to determine if code needs formal review or can remain read-only.
• Safety & Scrutiny:
  • Sandboxing: Vibe-coded apps run in isolated environments to prevent unauthorized external calls or credential misuse.
  • Formal Methods: A new team has been hired to implement formal methods to mathematically prove program properties for high-stakes areas.

Hiring & Organizational Culture

• Interview Process Differences (Palantir vs. Jane Street):
  • Palantir (Facet-Based): Focused on hiring "spiky" specialists; used distinct interviews for specific skills (e.g., "Learning," "Reengineering," "Decomposition") with single measurements per facet.
  • Jane Street (Ensemble-Based): Targets generalists; uses multiple similar interviews with two interviewers to build confidence across broad skill sets (coding, communication, calibration of confidence).
  • Casale's Critique: The Palantir model risks false negatives on generalists; Jane Street risks missing "spiky" experts in areas like performance engineering or front-end specialization.
• Proposed Hiring Reform:
  • One "Showcase" Interview: Casale proposes adding at least one adaptive interview question for every candidate (especially experienced hires) to allow them to demonstrate deep specialization (e.g., a "CTO-y" architectural problem or a specific performance challenge).
  • Rationale: To capture high-value talent that performs "just fine" in generalist assessments but excels in specific domains.
• Flexibility & Reorganization:
  • Internal Mobility: Jane Street favors small, continuous reorganizations to keep people on high-impact work; engineers often pivot between teams.
  • Technology Agnosticism: Maintains a limited language stack (OCaml, Python, C++) to maximize employee flexibility and prevent "pigeonholing" into specific tools.
  • Specialization Recognition: Acknowledges the high value of hardware and front-end engineers, despite industry stigma, and seeks to hire those with software engineering mindsets who specialize in these areas.