Designing a Robust LLM-Based Evaluation System for Agentic AI in Drug Discovery Through Human Alignment

• URL: http://arxiv.org/abs/2608.21057v1
• Date: Fri, 21 Aug 2026 12:58:24 GMT
• Title: Designing a Robust LLM-Based Evaluation System for Agentic AI in Drug Discovery Through Human Alignment
• Abstract summary: We present an LLM-as-a-Judge evaluation framework for ChatInvent, an agentic drug discovery assistant deployed at AstraZeneca.<n>We validate the judge through a human alignment study with five expert annotators, comparing Gemini 3.1 Pro, Claude Opus 4.7, GPT-5, and Llama 3.1 70B as candidate judges.<n>Third, we optimize the best-performing judge using few-shot demonstrations of human-annotated examples, improving alignment with the human majority vote from 0.80 to 0.86.
• Score: 0.764671395172401
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Agentic large language model (LLM) systems are reshaping scientific workflows in chemistry and drug discovery, but evaluating their open-ended, tool-augmented outputs remains a fundamental bottleneck. Reference-based metrics such as BLEU and ROUGE fail to capture semantic correctness, while expert human evaluation does not scale to the iteration speed these systems demand. The LLM-as-a-Judge paradigm has emerged as a scalable alternative, but existing drug discovery benchmarks deploy LLM judges without validating their alignment with human experts. In this work, we present an LLM-as-a-Judge evaluation framework for ChatInvent, an agentic drug discovery assistant deployed at AstraZeneca, with four contributions. First, we define four output-quality evaluation dimensions---Completeness, Relevancy, Structural Clarity, and Scope Adherence---alongside deterministic Tool Call Correctness checks. Second, we validate the judge through a human alignment study with five expert annotators, comparing Gemini 3.1 Pro, Claude Opus 4.7, GPT-5, and Llama 3.1 70B as candidate judges. Third, we optimize the best-performing judge using few-shot demonstrations of human-annotated examples, improving alignment with the human majority vote from 0.80 to 0.86. Fourth, applying the optimized judge to 70 held-out questions, we surface concrete limitations and find that informal phrasings do not systematically degrade output quality; if anything, it is helpful to have the LLM rewrite the original question before querying the agent. Our framework provides a reusable template for human-aligned evaluation of agentic systems in scientific domains.

Related papers

• Benchmarking LLM Judges for Mobile Agent Evaluation [19.646382307148418]
  We introduce MobileJudgeBench, a benchmark for evaluating LLM-as-judge methods on mobile agent trajectories.<n>Our benchmark comprises 931 human-annotated trajectories spanning 6 mobile agent benchmarks, 4 agent models, and 68 apps.<n>Our experiments reveal three key findings. First, a simple baseline judge with sampled screenshots is competitive with, and often exceeds, purpose-built methods.<n>Second, benchmark quality metrics reliably predict real-world judge utility.
  arXiv  Detail & Related papers  (2026-08-11T21:00:46Z)
• Beyond Text Matching: Towards Reference-Free Evaluation for Human-Oriented Binary Reverse Engineering [30.567931362297553]
  We introduce BinJudgeBench, the first expert-annotated, reference-free evaluation benchmark based on multi-dimensional human judgment.<n>BinJudge improves correlation with human experts by 4.5%-24.7% and reduces API cost to 0.06$times$-0.84$times$ of that of static best configurations.
  arXiv  Detail & Related papers  (2026-08-07T09:48:25Z)
• Counsel: A Meta-Evaluation Dataset for Agentic Tasks [8.10022820678836]
  Counsel is the first public dataset of meta-evaluations for agentic tasks.<n>It consists of process-level critiques from open-weight LLMJs on two agent benchmarks.<n>Human annotators label critiques on each flagged error as "spot on", "correct location but poor reasoning", or "should not have flagged"
  arXiv  Detail & Related papers  (2026-06-19T17:30:56Z)
• Time to REFLECT: Can We Trust LLM Judges for Evidence-based Research Agents? [61.49434544687523]
  We introduce REFLECT, a meta-evaluation benchmark targeting fine-grained failure detection in agentic environments.<n> REFLECT defines a detailed taxonomy of process- and outcome-level failure modes, instantiated by performing controlled and localized interventions.<n>Our experiments show that even the best-performing models achieve overall accuracies below 55% across reasoning, tool-use, and report-quality failures.
  arXiv  Detail & Related papers  (2026-05-18T23:55:08Z)
• How Trustworthy Are LLM-as-Judge Ratings for Interpretive Responses? Implications for Qualitative Research Workflows [0.6437935154416734]
  This study examines whether LLM-as-judge evaluations meaningfully align with human judgments of interpretive quality.<n>We generated one-sentence interpretive responses using five widely adopted inference models.<n>Results show that LLM-as-judge scores capture broad directional trends in human evaluations at the model level but diverge substantially in score magnitude.
  arXiv  Detail & Related papers  (2026-03-09T15:22:41Z)
• Quantitative LLM Judges [60.773734899532336]
  We propose quantitative LLM judges, which align evaluation scores of existing LLM judges to humans in a given domain.<n>The models are trained to improve the score of the original judge using its rationale and score.<n>Our experiments show that quantitative judges can improve the predictive power of existing judges through post-hoc modeling.
  arXiv  Detail & Related papers  (2025-06-03T14:44:23Z)
• J4R: Learning to Judge with Equivalent Initial State Group Relative Policy Optimization [69.23273504123941]
  We train judges to be robust to positional biases that arise in more complex evaluation settings.<n>We introduce ReasoningJudgeBench, a benchmark that evaluates judges in diverse reasoning settings not covered by prior work.<n>We train Judge for Reasoning (J4R), a 7B judge trained with EIS-GRPO that outperforms GPT-4o and the next best small judge by 6.7% and 9%.
  arXiv  Detail & Related papers  (2025-05-19T16:50:35Z)
• J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning [54.85131761693927]
  We introduce J1, a reinforcement learning framework for teaching LLM judges to think before making decisions.<n>Our core contribution lies in converting all judgment tasks for non-verifiable and verifiable prompts into a unified format with verifiable rewards.<n>We then use RL to train thinking-judges at scales of 8B, 32B, and 70B and show that they obtain state-of-the-art performance.
  arXiv  Detail & Related papers  (2025-05-15T14:05:15Z)
• Multi-Agent LLM Judge: automatic personalized LLM judge design for evaluating natural language generation applications [0.0]
  Large Language Models (LLMs) have demonstrated impressive performance across diverse domains, yet they still encounter challenges such as insufficient domain-specific knowledge, biases, and hallucinations.<n>Traditional evaluation methods, which rely on word overlap or text embeddings, are inadequate for capturing the nuanced semantic information necessary to evaluate dynamic, open-ended text generation.<n>We propose a novel dynamic multi-agent system that automatically designs personalized LLM judges for various natural language generation applications.
  arXiv  Detail & Related papers  (2025-04-01T09:36:56Z)
• SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal [64.9938658716425]
  SORRY-Bench is a proposed benchmark for evaluating large language models' (LLMs) ability to recognize and reject unsafe user requests.<n>First, existing methods often use coarse-grained taxonomy of unsafe topics, and are over-representing some fine-grained topics.<n>Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations.
  arXiv  Detail & Related papers  (2024-06-20T17:56:07Z)
• Calibrating LLM-Based Evaluator [92.17397504834825]
  We propose AutoCalibrate, a multi-stage, gradient-free approach to calibrate and align an LLM-based evaluator toward human preference. Instead of explicitly modeling human preferences, we first implicitly encompass them within a set of human labels. Our experiments on multiple text quality evaluation datasets illustrate a significant improvement in correlation with expert evaluation through calibration.
  arXiv  Detail & Related papers  (2023-09-23T08:46:11Z)
• Large Language Models are Not Yet Human-Level Evaluators for Abstractive Summarization [66.08074487429477]
  We investigate the stability and reliability of large language models (LLMs) as automatic evaluators for abstractive summarization. We find that while ChatGPT and GPT-4 outperform the commonly used automatic metrics, they are not ready as human replacements.
  arXiv  Detail & Related papers  (2023-05-22T14:58:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.