{"id":47239982,"title":"Show HN: Endgame – Production-aware ML under the sklearn API","points":1,"user":"cameronhamilton","time":1772576840,"time_ago":"10 hours ago","type":"link","content":"

Most ML frameworks optimize for leaderboard accuracy.\nBut in finance and healthcare, accuracy is often the least interesting part of the system.\nIf you can’t explain a prediction, you can’t deploy it.\nIf your probabilities aren’t calibrated, you can’t trust them.\nIf your pipeline doesn’t enforce constraints, you can’t ship it.\nI built Endgame after repeatedly running into that gap in production.

Anti-money laundering (banks)\nEarly in my career, I was hired to improve an anti-money laundering system.\nThe incumbent model was 28 hard-coded rules.\nIf enough thresholds fired (e.g., $3,000 ATM withdrawals over 30 days), the account was flagged.\nNo one knew where the thresholds came from.\nThere was no modeling of the underlying behavior.\nJust rule accumulation.\nI convinced the bank to provide the raw financial features behind those rule firings.\nWe trained an interpretable ML model directly on the underlying activity patterns.\nThe result: ~200% more true positives (accounts actually involved in fraud or laundering).\nBut what leadership cared about most wasn’t the metric.\nIt was this:\n“Why is this account suspicious?”\nThat theme repeated across industries.

Insurance claim adjudication\nI later built a claim adjudication model for a major health insurer.\nThe legacy system was massive, brittle, and effectively a black box.\nIt would frequently deny claims incorrectly, and no one fully understood how it worked.\nWe built a new ML system that brought claim-level adjudication accuracy to ~95%.\nAgain, the metric wasn’t the headline internally.\nThe headline was:\n“Why did this claim get denied?”\nIn regulated environments, interpretability isn’t optional.

Stock forecasting and calibration\nI also learned this lesson personally.\nI built stock-forecasting models that performed well in historical backtests.\nSome predictions showed 80% probability of a price increase.\nThen the market regime shifted.\nThe probabilities were overconfident.\nSome trades went the opposite direction.\nI lost money.\nAccuracy ≠ trustworthy probabilities.\nCalibration and drift awareness matter far more in deployment than most tutorials suggest.\nThat experience fundamentally changed how I think about ML systems.

The core idea\nEndgame is my attempt to encode those lessons into a framework.\nIt’s not trying to replace scikit-learn.\nEvery estimator implements fit / predict / transform.\nBut it extends the ecosystem with:\nGlass-box models (EBM, GAM, CORELS, SLIM, GOSDT, etc.)\nSOTA deep tabular models (FT-Transformer, TabPFN, SAINT, etc.)\nConformal prediction and Venn-ABERS calibration\nDeployment guardrails (leakage detection, latency constraints, drift checks)\n42 self-contained HTML visualizations\nSuper Learner, BMA, cascade ensembles\nA full AutoML pipeline that respects deployment constraints\nAll under a unified sklearn-compatible API.

Agent-native ML (MCP)\nWe’re in the agentic AI era.\nYou can ask an LLM to build a pipeline for you, but it often requires multiple prompts and manual corrections.\nEndgame ships with a native MCP server.\nThis lets agents:\nload data\ntrain models\ncompare results\ngenerate reports\nexport reproducible scripts\nThrough structured tool calls, not fragile prompt chains.\nMy belief is that ML pipelines will increasingly become conversational infrastructure.

A small contrarian view\nThe ML community is underestimating the problems left to solve in tabular data\nand overestimating the demand for accuracy-optimized models.\nMost real-world data in business, healthcare, and finance is tabular (often multimodal).\nAnd most real-world systems need to be interpretable, calibrated, and deployable — not just accurate.

Endgame v1.0.0 is open source (Apache 2.0).\nPython 3.10+.\nIf you work on production ML systems, especially in regulated domains, I’d genuinely value feedback.\nGitHub: https://github.com/allianceai/endgame\nInstall: pip install endgame-ml\nHappy to answer technical questions.","url":"item?id=47239982","comments":[],"comments_count":0}