Login or sign up
For business

Dmitry Krutous

4.4· 5 reviews
Average ratingGreat
4.4/ 5

Based on 5 reviews

560%
420%
320%
20%
10%
Ask AI

Questions about Dmitry Krutous

Based on 5 reviews, listing data and website checks · updated 8 days ago

What is Dmitry Krutous?

Dmitry Krutous is an ETF and asset-management domain expert based in the Greater Boston area who specializes in LLM evaluation, AI governance, and GenAI model validation for finance. He builds runnable evaluations that measure whether an AI can perform specific asset-management tasks using rubric-based grading with automatic failure gates. He is the author of TieOutBench, an open benchmark suite containing six evaluations of real asset-management workflows released under an MIT license.

What specific benchmark suite does the business provide?

The business offers TieOutBench, an open benchmark suite that includes evaluations for ETF creation/redemption reconciliation and OTC swap confirmation matching. Each evaluation decomposes workflows into scored checkpoints, grades against a point-weighted rubric with auto-fail gates, and runs live models end to end. The suite comprises 6 evaluations, 78 scored checkpoints, around 420 rubric criteria, and 20 gold cases.

What specific asset-management workflows are evaluated?

The evaluations cover back-office controls such as digesting 10-Qs for quarterly earnings analysis and recomputing cap and buffer from option strikes for defined-outcome ETF diligence. Other tasks include DCF valuation, corporate-actions processing, and NAV oversight for leveraged ETFs. These are described as the only public, runnable LLM evaluations of capital-markets post-trade operations as of October 2026.

How did the models perform in the October 2026 evaluations?

The options-redemption evaluation ran on eight frontier models from three vendors in October 2026, where every model held the broken redemption and settled the clean one. The corporate-actions evaluation also runs as an agent environment with tools like a document store and a test ledger. With tools, 35 of 48 runs passed every check, compared to 32 with documents in a single prompt.

How can users verify the reproducibility of the evaluations?

The benchmark suite is released under an MIT license, making it open source. The website states that everything above is reproducible using the command python -m harness demo, which runs the core with no API key. This allows users to verify the results without needing a paid account or key.

Where is the business located and how can it be contacted online?

The business is located in the Greater Boston area and serves clients interested in LLM evaluation and AI model validation for asset management. The website domain is evals.finance and the official site is https://evals.finance. The listing on Trustburn has a rating of 4.4 out of 5 from 5 reviews.

What are some of the most complex tasks evaluated?

The evaluations include defined-outcome (buffer) ETF diligence, which recomputes cap and buffer from option strikes and prices the protection. Another key feature is the ability to tie an authorized participant's basket to the published PCF for ETF creation/redemption reconciliation. These tasks require precise matching of financial data to ensure accuracy.

How does this business differ from other finance benchmarks?

The business focuses on back-office controls rather than front-office analyst work, distinguishing it from other benchmarks like GDPval or Vals AI's Finance Agent. It also runs as an agent environment with a second AI as a checker in two further arms. This dual-checker approach helps validate the accuracy of the initial model's output.

What are the practical requirements to get started with the tools?

The harness demo can be run with no API key, making it accessible for immediate testing. The website provides a footer link to Greater Boston, confirming the physical location. The MIT license ensures that the benchmark suite remains freely available for academic and commercial use.

What mechanisms ensure the reliability of the evaluations?

The benchmark suite includes 20 gold cases to ensure accuracy and reliability. The evaluations are designed to measure whether an AI can be trusted to do a specific finance job, graded like an experienced professional would grade work. This includes automatic failures for errors that quietly ruin an answer.

How does the system handle complex data scenarios?

The evaluations are designed to handle distractors and amendments in corporate-actions processing, ensuring the correct version is pinned. The system also computes entitlements based on deadlines, committing only what is allowed. This level of detail is crucial for accurate financial modeling.

What are the key features for developers?

The business provides a way to test the core functionality without any API key, making it easy for developers to integrate. The website explicitly states that the harness demo is reproducible, encouraging transparency. The MIT license further supports open collaboration and adoption.

What are the key features for finance professionals?

The evaluations cover a wide range of tasks from digesting 10-Qs to recomputing NAV for leveraged ETFs. The system also includes a checker of the same model or a stronger checker to validate results. This multi-layered approach ensures high accuracy in financial evaluations.

What is the official website of Dmitry Krutous?

The official website of Dmitry Krutous is evals.finance.

How is Dmitry Krutous rated on Trustburn?

On Trustburn, Dmitry Krutous has a rating of 4.4 out of 5 based on 5 reviews.

AI-generated from reviews on this page. Verify anything important with the company directly.

David Chen
Reviews 1

It is fantastic to see such a comprehensive suite of evaluations tailored specifically for asset management. Covering everything from DCF valuation to OTC swap confirmation matching is exactly what the industry needs. The MIT license makes it accessible, and the live model leaderboard provides great transparency. This is a fantastic resource for anyone serious about AI governance.

Sarah Jenkins
Reviews 1

The TieOutBench approach offers a sophisticated way to validate AI models in asset management by focusing on specific workflows rather than blended accuracy. With six evaluations covering areas like quarterly earnings analysis and corporate-actions processing, the service provides granular insights into model performance. The inclusion of forty-two rubric criteria and twenty gold cases creates a rigorous testing environment. However, the distinction between front-office and back-office evaluations is notable, as other benchmarks primarily grade analyst work. This focus on post-trade operations fills a specific gap. While the technical report and GitHub repository provide transparency, potential clients should verify if the current model coverage aligns with their vendor requirements before committing to the validation process.

Elena Rostova
Reviews 1

The detailed breakdown of the six evaluations covering earnings analysis and ETF diligence is impressive. Each workflow is decomposed into scored checkpoints with a point-weighted rubric. This structured approach ensures that AI models are graded like experienced professionals, making the service highly effective for validating specific finance jobs before deployment.

Marcus Thorne
Reviews 1

Reading through the TieOutBench details, it is clear that Dmitry Krutous offers a robust solution for asset management AI validation. The ability to run six specific evaluations with over seventy scored checkpoints gives firms real confidence. Knowing that errors like wrong valuation methods are caught by auto-fail gates makes this service highly available and reliable for daily operations.

News

No news yet. This is where the business shares its own updates and press.

Photos

No photos yet. Photos shared by reviewers will show up here.

Widgets

Boost your business trust by displaying a widget on your website — a simple way to convert your visitors into customers.

Get the widget for your site →

About Dmitry Krutous

Dmitry Krutous is an ETF and asset-management domain expert based in the Greater Boston area who works on LLM evaluation, AI governance, and GenAI model validation for finance. He builds runnable evaluations that measure whether an AI can perform specific asset-management tasks, using rubric-based grading with automatic failure gates. He is the author of TieOutBench, an open benchmark suite containing six evaluations of real asset-management workflows, including ETF creation/redemption reconciliation and OTC swap confirmation matching. The benchmark comprises 6 evaluations, 78 scored checkpoints, around 420 rubric criteria, and 20 gold cases, released under an MIT license with a technical report, an SSRN posting, a GitHub repository, and a live model leaderboard.