Yagiz Ihlamur

Anirudh Jaidev Mahesh, Ben Griffin, Fuat Alican et al.

Large language models (LLMs) are increasingly used for high-stakes decision-making, yet existing approaches struggle to reconcile scalability, interpretability, and reproducibility. Black-box models obscure their reasoning, while recent LLM-based rule systems rely on per-sample evaluation, causing costs to scale with dataset size and introducing stochastic, hallucination-prone outputs. We propose reframing LLMs as code generators rather than per-instance evaluators. A single LLM call generates executable, human-readable decision logic that runs deterministically over structured data, eliminating per-sample LLM queries while enabling reproducible and auditable predictions. We combine code generation with automated statistical validation using precision lift, binomial significance testing, and coverage filtering, and apply cluster-based gap analysis to iteratively refine decision logic without human annotation. We instantiate this framework in venture capital founder screening, a rare-event prediction task with strong interpretability requirements. On VCBench, a benchmark of 4,500 founders with a 9% base success rate, our approach achieves 37.5% precision and an F0.5 score of 25.0%, outperforming GPT-4o (at 30.0% precision and an F0.5 score of 25.7%) while maintaining full interpretability. Each prediction traces to executable rules over human-readable attributes, demonstrating verifiable and interpretable LLM-based decision-making in practice.

Yagiz Ihlamur, Ben Griffin, Rick Chen

Structured launch signals on Product Hunt contain statistically significant predictive information for Series A funding outcomes. We construct PHBench from 67,292 featured Product Hunt posts spanning 2019-2025, linked to Crunchbase funding records via deterministic domain matching, identifying 528 verified Series A raises within 18 months of launch (positive rate: 0.78%). Our best-performing model, a three-component ensemble (ENS_avg, ENS_ISO, XGB) selected by validation F0.5, achieves F0.5 = 0.097 and AP = 0.037 (95% CI: 0.024-0.072; 4.7x lift over random) on the private held-out test set (103 positives). A paired bootstrap confirms a statistically credible advantage over the logistic regression baseline (AP delta: +0.013, 95% CI: [0.004, 0.039], p < 0.001; F0.5 delta: +0.056, 95% CI: [0.006, 0.122], p = 0.016). Validation-set metrics (F0.5 = 0.284, AP = 0.126) reflect best-of-144 selection bias on 53 positives and are reported for benchmark reproducibility only. We further evaluate three zero-shot Gemini models (Gemini 2.5 Flash, Gemini 3 Flash, and Gemini 3.1 Pro) in an anonymized numerical setting. The best LLM achieves AP = 0.034 (Gemini 3 Flash), below the LR baseline AP of 0.044. Notably, the most capable Gemini variant (Gemini 3.1 Pro, AP = 0.023) performs worst -- an unexpected pattern that warrants further investigation across providers and prompting strategies. Both ML and LLM models show the same temporal performance decay tracking the 2020-2021 funding boom and subsequent contraction, confirming the dataset captures genuine market structure rather than noise. PHBench provides a reproducible framework comprising public training, validation, and blind test splits; 61 engineered features; a five-metric evaluation harness; and a public leaderboard at https://phbench.com. All code, baseline models, and anonymized dataset splits are publicly available.

Maximilian Westermann, Ben Griffin, Aaron Ontoyin Yin et al.

Feature discovery from complex unstructured data is fundamentally a reasoning problem: it requires identifying abstractions that are predictive of a target outcome while avoiding leakage, proxies, and post-outcome signals. With the introduction of ever-improving Large Language Models (LLMs), our method provides a structured method for addressing this challenge. LLMs are well suited for this task by being able to process large amounts of information, but unconstrained feature generation can lead to weak features. In this work, we study reasoning control in LLMs by inducing cognitive behaviors for improving feature discovery. We introduce CoFEE (Cognitive Feature Engineering Engine), a reasoning control framework that enforces cognitive behaviors in how the LLM reasons during feature discovery. From a machine learning perspective, these cognitive behaviors act as structured inductive biases over the space of candidate features generated by the model. These behaviors have been exploited with success in ML models, and include backward chaining from outcomes, subgoal decomposition, verification against observability and leakage criteria, and explicit backtracking of rejected reasoning paths. In a controlled comparison, we show that enforcing cognitive behaviors yields features with higher empirical predictability than those under unconstrained vanilla LLM prompts. CoFEE achieves an average Success Rate Score that is 15.2% higher than the vanilla approach, while generating 29% fewer features and reducing costs by 53.3%. Using held-out feature evaluation, we assess whether cognitively induced features generalize beyond the data used for discovery. Our results indicate that, in our evaluated setting, reasoning control is associated with improvements in quality and efficiency of LLM-based feature discovery.

Yagiz Ihlamur

Predicting startup success from founder career data is hard. The signal is weak, the labels are rare (9%), and most founders who succeed look almost identical to those who fail. We engineer 28 structured features directly from raw JSON fields -- jobs, education, exits -- and combine them with a deterministic rule layer and XGBoost boosted stumps. Our model achieves Val F0.5 = 0.3030, Precision = 0.3333, Recall = 0.2222 -- a +17.7pp improvement over the zero-shot LLM baseline. We then run a controlled experiment: extract 9 features from the prose field using Claude Haiku, at 67% and 100% dataset coverage. LLM features capture 26.4% of model importance but add zero CV signal (delta = -0.05pp). The reason is structural: anonymised_prose is generated from the same JSON fields we parse directly -- it is a lossy re-encoding, not a richer source. The ceiling (CV ~= 0.25, Val ~= 0.30) reflects the information content of this dataset, not a modeling limitation. In characterizing where the signal runs out and why, this work functions as a benchmark diagnostic -- one that points directly to what a richer dataset would need to include.