Two decades of pattern
recognition research.
One infrastructure question.
Why does every quant trader and AI tool builder have to compute historical market context from scratch? Konseki was built to answer that — pre-computed, structured, updated daily.
Background
Songyang Guo's interest in market pattern recognition began in 2006 during PhD research in Econometrics at Monash University, where his work focused on stock market similarity and computational finance — the mathematical foundations of what Konseki does today. In 2008, while still completing his research, he joined Argus Capital Management as a quantitative trader, applying that research directly in live markets. He discontinued the programme in 2011 as his trading career accelerated.
He later founded Enigma Capital Management, leading the research and development of pattern-recognition trading systems. Across his hedge fund career, Songyang managed more than US$120 million across equities and global futures markets.
Over the past decade, Songyang has worked as a software engineer and technology entrepreneur while continuing to trade actively using proprietary pattern-recognition models. The infrastructure problem he kept running into — the time and compute cost of building historical context from scratch every time — became the product idea behind Konseki.
Monash University
Argus Capital Management
Technology Entrepreneur
Why Konseki
The infrastructure problem that led to Konseki is simple: building historical market context from scratch is expensive. Sourcing 20+ years of OHLC data to search against, engineering the right features, running cross-market pattern matching, computing outcome distributions, scoring reliability — done properly, that's weeks of infrastructure work before you get to the first useful output.
And even if you build the infrastructure once, you can't run it at query time. Cross-market pattern matching and clustering algorithms across hundreds of symbols and 20+ years of daily data are computationally intensive — a single symbol's full analysis can take hours depending on the algorithm and history length. For backtesting, that cost compounds catastrophically: the computation must repeat for every single day in the backtest window, turning one round of validation into weeks or years of execution time. For an AI trading assistant, the problem is different but equally fatal — users expect responses in seconds, not hours.
Every quant who wants to validate a strategy idea against historical conditions faces this overhead. Every AI tool builder who wants to give their model genuine market context faces the same barrier.
"The computation itself isn't novel. It's just impossible to run on demand. Pre-computation isn't an optimisation — it's the only architecture that makes this data usable at all."
The reason Konseki exists as infrastructure, not a tool.
Konseki solves that as infrastructure. The engine runs every close — currently across ~500 S&P 500 symbols, with broader universe coverage on the roadmap. 8 lookback periods. 4 forward windows, all nested in every response. Outcome distributions, reliability scores, match quality, MAE, MFE, and LLM-ready commentary — structured as JSON, updated daily, accessible via API. Every day's output is also stored as a permanent snapshot, so you can query any past date the same way you query today's — useful for backtesting against exactly what the engine would have shown on a given day, not just what it shows now. You consume the output. We handle the computation.
"What happened the last time markets looked like this?"
The question Konseki was built to answer — programmatically, at scale, every close.
The methodology is publicly documented. The outputs are publicly visible on the Coverage page. The free tier exists so you can evaluate the data quality before paying for it. The product stands on the evidence — not on claims about it.