An investing research platform built on the belief that elite portfolio management should be available to everyone.
Founded by George Dikos. Powered by independent research and AI.
The edge is process, not information
The tools used by the world’s best investors — systematic strategy testing, ensemble portfolio construction, AI-driven risk management — have never been inaccessible because of talent.
The cost and complexity barrier
They have been inaccessible because of cost and complexity.
Research and agentic AI close the gap
Goldflower’s thesis is that independent research coupled with agentic AI can make elite portfolio management available to the everyday investor — and exists to drastically remove these barriers.
Every day, the platform runs thirty stock screens — five index universes across three time windows and two list sizes — and publishes what each one selects, ranked by how many of the thirty hold each name. No single strategy. No gut feel. No hindsight: the screen set is fixed in advance.
The full methodology, and what it could and could not establish, is documented in the research paper. Read the research →
Strategy universe
Thousands of configurations across lookback, holding period, portfolio size and benchmark.
Pick the winners
Ranked point-in-time on data available that day. No future information, ever.
Pool and weight
Holdings from every winner are pooled; the better an asset ranks, the more weight it gets.
Execute at T+1
Signals trade at the next day’s close, so nothing can act on information it did not have.
1. Strategy universe
The engine enumerates strategy configurations across multiple dimensions: lookback windows (5 to 100 days), holding periods (1 to 20 days), portfolio sizes (5 to 30 assets), and benchmark indices. Each configuration produces a daily list of top-ranked assets based on historical performance patterns stored in a PostgreSQL database with 5+ years of market data.
2. Screen set (point-in-time)
The screen set is fixed in advance, so nothing is chosen with hindsight: five index universes, three lookback windows and two list sizes. At each evaluation date T only information available up to and including T is used, and point-in-time SQL queries enforce that.
= 30 fixed screens
3. Ranking by overlap
What each screen selects is pooled, and every stock is ranked by how many of the 30 screens hold it. The count is published as a fact about the screens rather than as a strength score: being held by six screens has not been shown to be better than being held by one. No weights are assigned.
Rank 1 (best) → highest weight
Assets appearing in multiple strategies use their best (lowest) rank
4. T+1 execution
All signals are executed at the next trading day’s close price. This strict rule eliminates lookahead bias entirely. The portfolio cannot trade on the same day a signal is observed.
Return measured from T+1 close to holding period end
5. Evidence and testing
What the screens do is measured directly: they select stocks that have just moved sharply. Whether those stocks then outperform a comparable basket is the subject of a three-year test, registered before its data existed, which began in October 2026 and reports in 2029. The evidence page carries the figures, including the matched comparison against similar stocks and what a sample of this length is able to detect.
Better portfolios and better risk management through the combination of systematic research and agentic AI — available to everyone, transparent, and improving every day.
Daily Ensemble Portfolio
Every morning, the engine publishes the stocks its screens select, ranked by how many of the 30 screens hold each one, updated daily and delivered before the market opens.
Agentic Risk Management
An AI layer screens holdings, detects regime shifts, weatherproofs portfolios, and helps execute — moving from recommendation to action.
A Learning System
Every recommendation is tracked. Every outcome measured. The goal is a system that gets measurably better every day — eventually managing portfolios autonomously, with full transparency.
Everyone Gets the Same Engine
No tiers, no better answer for a larger balance. The person with a first thousand invested sees exactly what everyone else sees, at the same time each morning.
I design, fund, and build Goldflower. My background is in structural econometrics, operational research, and applied machine learning. I was the first research scientist at Amazon EU, and before that I built integrated planning models for industrial operations at Heracles/Lafarge.
The idea started in 2014 after Tucker Balch’s quantitative investing course on Coursera and reading More Money than God. I started trading warrants of Greek banks (long) versus the underlying shares (short), proving to myself that long-short strategies could generate alpha with less volatility — and that this contradicted the efficient market hypothesis I’d studied at MIT. From that point I kept adding strategy types, introduced a pruning approach to shortlist the ones that worked, and configured portfolios from the winners. That system became EventSim, and four years of building it became Goldflower.
The name comes from my family. Like Greek ship owners who name their ships after people they love, I wanted to signal how personal this is. Goldflower is named after Chrysanthi — the golden flower.
AWS EC2 + PostgreSQL
Strategy engine and data store
Strategy Engine
Thousands of daily configurations · Python + yfinance
Daily Portfolio
Ensemble of winning strategies · AWS S3
Delivery
Daily email before market open · AWS Lambda + SES
Execution
Paper and live trading · IB Gateway
Intelligence
Screening, risk, conversation · Claude AI Agent