Thanks to visit codestin.com
Credit goes to github.com

Skip to content
View mkzung's full-sized avatar
💭
Rockin
💭
Rockin

Block or report mkzung

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mkzung/README.md

Max Gorbuk

I check whether published numbers are true. Most of what I find looks fine and is not: trading volume that no transfer supports, a price copied from another venue, a company valuation that nobody has tested since its last funding round.

Studying for the Master in Analytics & Management at London Business School. Research Analyst at the Stanford GSB Venture Capital Initiative under Prof. Ilya Strebulaev.

What I found

Tokenized stocks trade both on an exchange and in on-chain pools. The exchange sets the price and the pools follow, on 21 of the 23 pair-days I could rank across 24 tokens.

Of 73 low-cap pools screened on Base and BNB Chain, 10 flag on wash-trading mechanics and 9 are sustained, counted from the transfers rather than from an aggregator. Three more signatures are in the DN Institute wiki, and the same mints, screened on a central book and in Solana pools, are in tokenized-equity-wash-trading.

Mutual funds have to report a value every month for the private companies they hold, even when nothing has traded. unicorn-valuation-disagreement measures how far apart different managers value the same company on the same day, from SEC filings only. SSRN working paper.

Open source

89 pull requests merged in 18 organisations, most of them in ccxt, QuantConnect/Lean and the UK AI Security Institute's inspect_evals, and 15 more organisations.

It is usually the same bug. A project writes down a rule and never checks it, or a shared test runs against a list that is missing half the classes it should cover. Feed it something degenerate and the answer comes back finite and wrong.

I wrote the Inspect AI implementation of Do-Not-Answer, which UK AISI's inspect_evals lists, with an adversarial suite registered beside it. lm-refusal-eval measures the same thing and re-runs byte-identically from the same seed.

Elsewhere

Named contributor on the WEF and Stanford GSB report The Future of Venture Capital (2026). Founded MedAI, Rospatent-registered clinical decision support for paediatrics, running in ten clinics.

Python, pandas and Polars, Solana and EVM data, Inspect AI, some Rust and C++.

gorbuk.com · LinkedIn · Stanford · [email protected]

Pinned Loading

  1. ethbtc-suspicious-patterns ethbtc-suspicious-patterns Public

    Six-detector forensic analysis of ETH/BTC market data — wash-trading investigation with defensible, reproducible evidence. DN Institute Challenge #492 submission (PR 1712n/market-data-challenge#24).

    Python 1

  2. contract-tracker contract-tracker Public

    Email-driven contract-approval tracker — IMAP poller + RFC-5322 thread reconstruction + OpenAI JSON-mode classifier + Streamlit dashboard. Built for a construction firm.

    Python 1

  3. fundarb fundarb Public

    Cross-venue funding-rate arbitrage CLI for crypto perpetuals — Hyperliquid + Orderly + Backpack. Python, ed25519/EIP-712 signing, testnet/mainnet config, live signal generation + execution.

    Python 1 1

  4. research-automation-pipeline research-automation-pipeline Public

    Reference architecture for AI research-automation pipelines (n8n + Python + LLM) — generic template, derived from work at Stanford GSB Venture Capital Initiative, published with advisor permission.

    Python 1

  5. awesome-clinical-ai-safety awesome-clinical-ai-safety Public

    A curated list of resources on clinical AI safety: evaluation, regulation, deployment, and post-market surveillance for medical AI.

    1 1

  6. morpho-vault-counterfactuals morpho-vault-counterfactuals Public

    Historical replay + counterfactual stress testing for Morpho MetaMorpho vaults: 6 risk detectors (oracle freeze, collateral cascade, depositor exit, utilization, liquidation latency, LTV stress)

    Python 2