Research, engineering deep dives, and perspectives on AI-driven quantitative finance — from the team building ARKRAFT.
How we built a transformer-based regime detector that identifies market state transitions 2-3 days earlier than traditional Markov-switching models, and what it means for signal timing.
A deep dive into the autonomous anomaly detection and recovery system that keeps ARKRAFT's data pipelines running — including the Korea equity incident that resolved itself in 14 seconds.
Our annual review of how AI is reshaping quantitative finance — from autonomous signal discovery to the emerging role of agent-based systems in portfolio management.
How we built the EARNINGS_SENTIMENT_DIVERGENCE signal — a detailed look at the NLP pipeline, the divergence metric, and why management tone vs. analyst tone predicts short-term returns.
How we reduced average slippage from 1.2 bps to 0.3 bps by training an RL agent to dynamically select execution venues based on real-time liquidity conditions.
Lessons from 14 months of running multi-agent systems in production — what works, what fails, and why agent-to-agent trust is the hardest problem.
A summary of everything we built in Q4 2025 — including Decision Traces, the self-healing pipeline upgrade, and 12 new alternative data connectors.