Hi there! I work across machine learning, applied AI, and experimentation, with experience in recommender systems, NLP, computer vision, and generative AI.

LinkedIn · Google Scholar

• reasoninglab — Studies when test-time reasoning improves LLM performance versus when it simply adds compute. Chain-of-Thought · Test-Time Compute · Open-Weight LLMs · Ollama

• adaptlab — Compares prompting, RAG, LoRA, and their combination for adapting LLM behavior and knowledge. LoRA · RAG · Prompting · McNemar Test

• detectivelab — Tests how multimodal models reason over visual, structured, and conflicting evidence under uncertainty. Vision-Language Models · Multimodal Reasoning · Selective Prediction · Uncertainty Estimation

• evallab — Studies whether observed LLM performance differences are statistically trustworthy and reproducible. Paired Evaluation · Bootstrap CI · Benchmark Stability · Experimental Design

• SearchRankLab — Evaluates sparse, dense, hybrid, and reranked retrieval across relevance, cost, and domain-shift tradeoffs. BM25 · Sentence Transformers · Reciprocal Rank Fusion · Cross-Encoder

• PersonalRankLab — Studies when recent user behavior improves personalized ranking and when recency weighting hurts compared with full-history user representations. Personalization · Temporal Modeling · NDCG@10 · User-Clustered Bootstrap

• CommerceRecLab — Builds a session-based e-commerce recommender spanning candidate generation, ranking, cold start, and serving tradeoffs. Session-Based RecSys · Implicit Feedback · Candidate Generation · NDCG@20

• ReciprocalMatchLab — Models two-sided preferences and mutual-match ranking with calibrated probabilities and exposure-aware allocation. Logistic Regression · Probability Calibration · NDCG · Brier Score

• PolicyRecLab — Studies off-policy evaluation and exploration for contextual-bandit recommendation under biased logged data. Contextual Bandits · IPS / SNIPS · Doubly Robust · Propensity Scores

• PolicyShiftLab — Studies how policy-induced selective observation changes offline recommender evaluation and model selection compared with randomized or target evaluation. Offline Evaluation · Covariate Shift · IPW / SNIPS · Brier Score

• ObserveLab — Studies how serving policies shape what recommenders can learn and when exploration offsets exposure bias. Thompson Sampling · Contextual Bandits · Exploration–Exploitation · Exposure Bias

• ranklab — Tests how logging policy and behavioral targets affect offline recommender model selection. BPR · LightGCN · NDCG@10 · Randomized Logging