Introducing Heurismatic
Software that learns
to improve itself.
Every program is full of hard-coded rules written by humans.
Heurismatic finds them, learns from them, and replaces them with something better.
Introducing Heurismatic
Every program is full of hard-coded rules written by humans.
Heurismatic finds them, learns from them, and replaces them with something better.
The Problem
Developers write rules based on intuition. "If the server is busy, add more capacity." "If the price dropped, sell." "If the queue is long, skip ahead."
These rules are rigid. They don't adapt to new patterns. They can't learn from mistakes. And they're buried so deep in codebases that most teams don't even know they're there.
How It Works
Heurismatic crawls through your codebase like a search engine crawls the web. It automatically finds every place where a hard-coded rule is making a decision.
No manual setup. Point it at your code and it maps every decision point.
For each rule it finds, Heurismatic trains a tiny, lightweight model that watches how the rule behaves in the real world. Over time, the model learns to make better decisions than the original rule ever could.
Think of it as a student that observes the teacher, then surpasses them.
Heurismatic measures the improvement before making any changes. Only when the learned model outperforms the old rule does it swap in the replacement. Everything is validated automatically.
No guessing. Every replacement is backed by measured improvement.
Proven Results
Replaced the scaling heuristic in an open-source Kubernetes GPU autoscaler.
Replaced entry/exit rules in an open-source algorithmic trading framework.
The Simplest Way to Think About It
"Imagine every piece of software has a thousand tiny thermostats. Someone set them all by hand, years ago. Heurismatic replaces each one with a smart thermostat that adjusts itself."
Where It Applies
Autoscaling, load balancing, resource allocation. Replace static thresholds with models that predict demand.
Trading signals, risk thresholds, fraud detection rules. Models that adapt to changing market conditions.
Query planning, indexing strategies, cache eviction. Let the database learn from its own workload patterns.
Difficulty scaling, NPC behavior, matchmaking. Replace hand-tuned values with adaptive intelligence.
Production scheduling, quality thresholds, predictive maintenance triggers. Smarter decisions on the line.
Route optimization, inventory reorder points, delivery scheduling. Replace fixed rules with learned ones.
What Makes It Different
You don't need to know where the rules are. Heurismatic finds them for you by crawling your entire codebase.
Not locked to one domain. Cloud infra, trading systems, databases, games. If it has rules, Heurismatic can improve them.
Tiny models that run at the speed of the original code. No heavy GPU infrastructure needed. Deploys at the edge.
Every replacement is benchmarked. The old rule and new model run side by side. Only proven improvements go live.
Heurismatic is in early development. We're looking for teams with complex, rule-heavy codebases who want to be first adopters.
Get in Touch