A track record, not a portfolio of logos.

Institutional analysts would not trust an AI that could not show its math. Research platforms shipped in 2024 led with the chat interface and treated grounding as a feature; the result was answers analysts could not defend in an investment committee.

Built the verification layer before the chat interface — the opposite of how most teams shipped. Filings, transcripts, and live market data are ingested into a Financial Knowledge Graph. Queries route through agentic workflows on LangGraph (supervisor + swarm orchestration with deterministic handoffs) that verify every numerical claim against source tables before delivery, with grounding-chunk citations on every paragraph and contradiction detection across filings and timeframes. Per-firm policy is injected at the orchestrator layer; refusal triggers fire on low-grounding or stale sources.

The market offered bolt-on AI features stapled to decade-old gradebooks. The vision needed the opposite: an AI-native core with everything a school runs on — CRM, parent portal, curriculum authoring, LMS, AI tutor, grading, engagement analytics, and student evaluation — in one coherent system.

Built the entire platform in two months, deployed across three continents. The hard call was refusing to ship the tutor until it could cite the textbook page on every answer — source-constrained agentic RAG with zero-hallucination as a release gate. A multimodal Socratic grading engine handles text, image, and notebooks; real-time content moderation runs for K–12 student safety; a serverless stack keeps it sub-second across continents.

A multi-asset general account lacked a repeatable, governance-backed framework for dynamic allocation across regimes. Risk and return were managed separately from the decision rituals that let the team act when it counted.

Built quantitative frameworks for asset allocation and risk management with cross-asset factor analysis on the $80B portfolio, and established the Transaction Review Committee. War-gamed a macro downturn fourteen months before March 2020, then executed the playbook — deploying $1B across eleven trading days while the committee reconvened daily. Governance rituals, not models, are what let you act when it counts.

Inherited a fixed income book built for a regime that was ending. The existing structure could not express the views needed inside the firm's active-risk budget.

Redesigned the $5B fixed income book inside an active-risk budget to double expected IR, without flinching mid-transition. Managed a $240M U.S. yield-curve portfolio to an information ratio of ~2.0. Risk limits are a feature, not a constraint — most of how the practice thinks about portfolio construction today started at that desk.

Two separate problems: scoring decisions under extreme market stress through the financial crisis, and then running an inventory-dependent business with the rigor of a portfolio manager rather than a merchant.

At FrontPoint / Copia Capital, built equity factor models and a regime-based currency strategy that delivered three consecutive positive years at ~30% returns, plus $100M+ in alpha from beta timing — learning that factor regimes are macro-driven and the two cannot be separated. At Duli Capital, treated inventory the way a portfolio treats positions: position sizing, drawdown limits, and kill criteria. Profitable from month one because the model said no to most of what looked good. Capital discipline beats conviction, every time.