An LLM can say anything, which is exactly the problem. The same open-endedness that makes it useful means it can hallucinate, leak data, follow a malicious instruction hidden in its input, or produce something harmful — and in production, in front of real users, those aren't edge cases, they're inevitabilities. Guardrails are the controls that sit around the model to contain what it might do. This post covers the new failure modes LLMs introduce and the layered defenses that keep a production system safe.
An LLM can say anything, which is the problem. The new failure modes — hallucination, prompt injection, data leakage, harmful output — and layered, fail-closed guardrails (input, output, action) with humans in the loop, treated as part of the lifecycle and red-teamed.
Underwriting is the decision to do a deal; monitoring is everything after — watching whether the thesis still holds as the world changes, the borrower performs or deteriorates, and covenants tighten or breach. It's traditionally periodic, manual, and uneven, which means problems get noticed late. AI makes monitoring continuous: reading every update, tracking every covenant, and flagging variance from the original thesis the moment it appears. This post is about building the monitoring layer.
Monitoring is the long aftermath of a deal — watching whether the thesis still holds as the world changes. AI makes it continuous: reading every update, tracking every covenant, flagging variance from the thesis the moment it appears — attacking the classic failure of problems noticed too late. The make-or-break is signal over noise. (Educational, not financial advice.)
Derivatives sound exotic, but a derivative is just a contract whose value derives from something else — a stock, a rate, a commodity. What makes them intellectually beautiful is how they are priced: not by predicting the future, but by insisting that their price be consistent with the things they depend on. That insistence — the no-arbitrage principle — is the single most powerful idea in quantitative finance, and this post builds it from the ground up.
A derivative's value derives from something else — and it's priced not by prediction but by the no-arbitrage principle: if a portfolio replicates a payoff, it must cost the same. Forwards, futures, options, payoff shapes, and the replication argument that prices a forward and links calls to puts. (Educational, not financial advice.)
Offline evaluation tells you whether a change is safe to ship; online evaluation tells you whether it actually worked. No matter how good your eval set is, production always surprises you — real users send inputs you never imagined, and the gap between your offline scores and real-world quality is where LLM apps quietly fail. This post is about measuring quality on live traffic and turning user behavior into the feedback that makes the system improve.
Offline evals say a change is safe to ship; online evals say whether it worked. The intrinsic offline–online gap, deploying to measure (canary, A/B, shadow), capturing explicit and implicit feedback, and the pipeline that turns production behavior into continuous improvement.
Underwriting is where a lender or investor decides whether a deal is worth doing — reading the materials, assessing the risk, forming a thesis, and documenting the judgment for an investment committee. It's slow, manual, judgment-heavy work, and it's one of the highest-value places AI can help: turning raw deal materials into structured, pressure-tested, auditable analysis. This post is about building AI underwriting that accelerates the work without surrendering the judgment or the accountability.
Underwriting turns deal materials into a decision — slow, manual, judgment-heavy work and a flagship AI application. Turning raw materials into structured, pressure-tested, auditable analysis: reading the deal record, surfacing risks, pressure-testing against precedent, drafting IC docs — while judgment stays human and everything stays traceable. (Educational, not financial advice.)
You cannot have higher expected return without accepting more risk — that trade-off is the deepest law in investing, and quantifying it precisely is the achievement that launched modern finance. The surprising twist is that risk is not simply additive: combine imperfectly related assets and some risk cancels out for free. Diversification is the closest thing finance has to a free lunch, and portfolio theory is the math of exploiting it.
Higher expected return requires more risk — and the surprising twist is that diversification reduces risk for free when assets are imperfectly correlated. Volatility, covariance, the efficient frontier, systematic vs. idiosyncratic risk, and how CAPM prices risk through beta. (Educational, not financial advice.)
The single practice that separates teams who improve their LLM apps from teams who just change them is evaluation. Without a trustworthy way to measure quality, every prompt tweak is a gamble — you fix one case, break two others, and never know. Eval-driven development makes the eval set the center of gravity: you build it before you optimize, you gate every change on it, and you treat it as the most valuable asset the project owns. This post is about building evaluation you can actually trust.
The practice that separates improving an LLM app from merely changing it. The curated eval set as the core asset, how to score open-ended output (deterministic checks, reference metrics, LLM-as-judge, human review), and how to gate every change on evals and compare B vs A.
Finance runs on documents — filings, offering memos, indentures, contracts, credit agreements, research — and until recently machines couldn't read them in any useful way. Large language models changed that, and it's arguably the most consequential development in AI financial engineering: the ability to read, extract, and structure the vast unstructured corpus that finance has always been drowning in. This post is about turning financial documents into analyzable information, reliably and auditably.
Finance runs on documents — filings, memos, indentures, contracts — and LLMs can finally read them at scale. The most consequential capability in AI financial engineering, and the discipline it requires: grounding, provenance on every output, hallucination defense, and human verification. (Educational, not financial advice.)
A bond is the simplest real instrument in finance — a contract to pay fixed cash flows on fixed dates — which makes it the perfect place to turn the time value of money into something concrete. But bonds also introduce the two ideas that run through all of fixed income: yield, the single number that summarizes a bond's return, and the yield curve, the market's price of time itself. Understanding bonds is understanding how markets discount the future.
A bond is the simplest real application of discounting: fixed future cash flows, priced by present value. From there come yield, the inverse price-yield relationship, duration and convexity as interest-rate risk, and the yield curve as the market's price of time. (Educational, not financial advice.)
In an LLM application the prompt is the source code — it's where behavior is defined and where most changes and most bugs happen. Yet teams routinely paste prompts inline, edit them in production, and have no idea which version produced last week's output. Prompt and version management is the unglamorous discipline of treating prompts as the first-class, versioned, tested artifacts they actually are. It's the foundation the rest of the lifecycle stands on.
In an LLM app the prompt is the source code — where behavior is defined and most bugs happen. Treat prompts like code: version the whole configuration (template + model + parameters + context strategy), decouple them from app code, and route every change through review, eval, rollout, and rollback.
Predicting where markets go is the most attempted and most humbling problem in finance. Financial time series look forecastable — they have trends, cycles, and patterns — but they are among the hardest series to predict, for deep structural reasons. This post applies time-series forecasting to financial data honestly: what the special difficulties are, what can and can't be forecast, and why the goal is usually not "predict the price" but something more modest and more useful.
Predicting where markets go is the most attempted and humbling problem in finance. Why financial series are especially hard (near-efficiency, noise, non-stationarity, fat tails, reflexivity), what can actually be forecast (volatility over direction, distributions over points), and why the goal is risk, not alpha. (Educational, not financial advice.)
A practical blueprint for founders and GTM leaders: when to start forward deployment, who to hire first, how to scope engagements, what to measure, and how to keep it from eating your margins.
A practical blueprint for founders: when to start forward deployment, who to hire first, how to scope engagements, what to measure, and how to keep it from eating your margins — plus the four failure modes (agency, hero, science-project, over-serving) and a 90-day start.