InCUBIG Tech BlogbyHyungyu Lee·Just nowWhen Human Approval Becomes DataAn approval says a person did not object to an AI’s output. Filed as a human answer, it can end up in the answer key that grades the same AI
Ankit Goel·16h agoYour RAG Works in the Demo. Here’s Why It Fails in Production.Precision@K, Recall@K, MRR and NDCG, plus a 6-step trace to find where your RAG pipeline breaks. Part 3 of 3.
Navaneethan L·20h agoEvaluation Engineering: Building Evals That Actually WorkMost AI teams have evals. Most AI teams’ evals do not work.A response icon1A response icon1
Yashar Mehdad·23h agoEMNLP 2026: Models Are No Longer the Whole Product!What ~5k papers, production systems, and a strained review process reveal about the next phase of NLP.A response icon1A response icon1
InCrater LabsbyKhushboo Thaker·1d agoThe Illusion of Memory: What We Learned Evaluating Three LLM Memory SystemsKhushboo Thaker and Yony Bresler · Crater LabsA response icon1A response icon1
Amajuoyi Udochukwu Deborah·1d agoBuilding EvalSpree: an evals tool that ends in a launch decisionEvalSpree is a free evals tool that helps product managers decide when an AI feature is ready to ship. I designed and built it myself, as a…
Ndeto Martin·2d agoConsistent Agents: Evaluating Agent BehaviorIn traditional software, the same input produces the same output. This is deterministic. Large language models are not: even with low…
Chamath Palihawadana·2d agoJudge the trajectory, not just the outputAn AI agent says it checked your company’s documents before answering. The answer looks right. But did it actually check?
InAlt.byAkshay Bhardwaj·2d agoFrom LLMs to DPO: A Practical Guide to Training and Fine-Tuning Large Language ModelsUnderstanding pretraining, supervised fine-tuning, LoRA, QLoRA, RLHF, PPO, and DPO through real-world engineering applications.A response icon1A response icon1
Brahim Guaali·2d agoYour AI agent needs a referee, not better promptsPart 4 of 4 in Building an AI agent from first principles. Part 1 built a refund agent as a graph, part 2 took it to production and part 3…
