QuantDrill

Problem

Quant interview preparation is fragmented across textbooks (heard on the street), forums, and expensive coaching. The questions span probability, stochastic calculus, brain teasers, and coding — each requiring different preparation strategies. Existing platforms either focus on software engineering interviews (LeetCode) or are static question banks without adaptive difficulty.

Solution

QuantDrill is an AI-powered interview preparation platform purpose-built for quant finance roles. It generates questions across all quant interview categories, renders mathematical notation in real-time via KaTeX, evaluates solutions using LLM judges, and adapts difficulty based on candidate performance.

Key Technical Decisions

KaTeX for math rendering — Quant questions involve complex mathematical notation (integrals, stochastic processes, matrices). MathJax is the traditional choice but is heavy and slow for real-time rendering. KaTeX renders 10x faster with a smaller bundle, which matters when questions are generated dynamically and need to render instantly.

LiteLLM for multi-provider LLM routing — Question generation and evaluation are LLM-intensive. LiteLLM abstracts away provider differences, enabling automatic fallback between GPT-4o and Claude based on latency, cost, and availability. For math-heavy evaluations, Claude tends to be more accurate; for verbal reasoning, GPT-4o is faster. The router optimizes for each question type.

Celery for async evaluation — Some evaluations (especially multi-step proofs) take 10-15 seconds of LLM processing. Running these synchronously would block the UI. Celery workers process evaluations asynchronously, with Redis as the message broker and result backend. The frontend polls for completion, keeping the experience responsive.

Impact

• Covers all quant interview categories: probability, stochastic calculus, brain teasers, statistics, coding
• Real-time KaTeX mathematical rendering for complex notation
• Adaptive difficulty based on candidate performance history
• Multi-provider LLM evaluation via LiteLLM with automatic failover
• Community tool potential — similar to how Virat Singh's ai-hedge-fund built community credibility