In large-scale construction projects, the continuous evolution of decisions generates extensive records, most often captured in meeting minutes. Since decisions may override previous ones, professionals often need to reconstruct the history of specific choices. Retrieving such information manually from raw archives is both labor-intensive and error-prone. From a user perspective, we address this challenge by enabling conversational access to the whole set of project meeting minutes. Professionals can pose natural-language questions and receive answers that are both semantically relevant and explicitly time-annotated, allowing them to follow the chronology of decisions. From a technical perspective, our solution employs a Retrieval-Augmented Generation (RAG) framework that integrates semantic search with large language models to ensure accurate and context-aware responses. We demonstrate the approach using an anonymized, industry-sourced dataset of meeting minutes from a completed construction project by a large company in Belgium. The dataset is annotated and enriched with expert-defined queries to support systematic evaluation. Both the dataset and the open-source implementation are made available to the community to foster further research on conversational access to time-annotated project documentation.
Figures & tables
Figure 2: Dataset composition and temporal statistics. (a)-(c) Summarize document structure: page, passage, and word counts. (d) Illustrates the temporal distribution of documents over the project’s time span.
Figure 3: Evolution of a subject in the meeting minutes . The initial decision is amended over meetings and truncated for brevity toward the end of the project. Decision text translated from French.
Figure 4: System implementation overview . Schematic representation of the implemented system designed to meet the use case specifications. Dashed lines denote offline operations, while solid lines represent online processes. Modules enclosed within the gray dotted frame are optional and not essential for core system functionality.
nbatch
tretrieve
trerank
ttotal
P@5
R@5
F1@5
1
5.77
190.9
275.4
0.34±0.10
0.92±0.09
0.48±0.11
2
3.19
95.98
133.3
0.49±0.15
0.85±0.12
0.59±0.13
6
1.42
31.13
55.2
0.73±0.20
0.57±0.17
0.60±0.18
10
1.08
19.48
27.9
0.71±0.23
0.43±0.15
0.47±0.14
12
0.98
16.01
22.8
0.72±0.15
0.43±0.16
0.47±0.15
30
0.72
6.15
10.6
0.88±0.20
0.27±0.21
0.35±0.19
Table 1: Execution times and retrieval performance as a function of batch size nbatch . Times ( tretrieve , trerank , and ttotal ) are in seconds and averaged over 8 queries. Precision@5 (P@5), Recall@5 (R@5), and F1@5 metrics summarize the retrieval quality.
Figure 5: System Graphical User Interface . Screenshot of the application interface showing the bottom input box for user queries, the left panel with the decision timeline and system replies, and the right panel displaying retrieved source pages grouped by time span and document.
Figure 6: Query - document relevance heatmap . Heatmap showing the distribution of pages containing relevant passages for each query in the benchmarking set. Rows correspond to evaluated queries, and columns represent documents ordered chronologically (dates shown on the top axis). Color intensity indicates the number of relevant pages cited per document.
ttotal (s)
P@5
R@5
F1@5
Rerank:
Off
top-5
Off
top-5
Off
top-5
Off
top-5
10
8.61
16.16
0.68
0.74
0.40
0.46
0.45
0.50
± 0.19
± 0.23
± 0.11
± 0.14
± 0.10
± 0.15
top- k
20
12.28
26.28
0.66
0.73
0.39
0.45
0.44
0.49
± 0.20
± 0.22
± 0.09
± 0.14
± 0.10
± 0.14
50
24.18
57.38
0.69
0.72
0.42
0.41
0.46
0.47
Table 2: Comparison of runtime and retrieval performance for different top- k values . Batch size nbatch=10 . Precision, Recall, and F1 metrics include standard deviation (reported below the mean).
Figure 7: Retrieval performance metrics at query and global levels . Evaluation of retrieval metrics at different cutoff values keval=2,3,4,5 , comparing system performance across queries and in aggregate.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Figure 9: System output comparison for different batch sizes . Comparison of system-generated outputs for batch sizes nbatch=6,10,and 60 . Each sample reply illustrates how varying the batch size affects the completeness and consistency of the retrieved and synthesized responses.
Meeting archives are difficult to search when users remember what was discussed but not when. We study topic-to-timestamp alignment: given a natural-language topic and a timestamped meeting transcript, the goal is to return the time at which the topic is discussed. A standard RAG setup can retrieve relevant transcript excerpts, but still asks the language model to generate a timestamp, which can produce unsupported or invalid timecodes. We therefore recast timestamp prediction as constrained temporal candidate selection: the system retrieves timestamped transcript chunks, and the model selects the candidate that best grounds the topic instead of generating a timecode. On 420 topic-timestamp queries from 200 municipal meeting transcripts, this increases Recall@5 from 31.9% to 50.0%, reduces MAE from 837.0 seconds to 761.0 seconds with Mistral-7B-Instruct, and increases the number of parseable outputs from 373 to 419 of 420 queries. The results suggest that temporal grounding in long transcripts depends strongly on retrieval quality and output design, not only on the choice of the language model.
Zeynep Yılbırt, Marina Litvak, Michael Färber
TU Dresden, Germany · SpeechMind GmbH, Germany · Shamoon College of Engineering, Israel +1
Query-centric cross-document event timeline extraction aims to construct event timelines from multiple documents under diverse query focuses. The task is very challenging due to the heterogeneous event modeling and reasoning requirements arising from both the intrinsic complexity of cross-document timeline construction and the diversity of query focuses. In this paper, we propose TimelineAgent, an LLM-based multi-agent framework featuring a multi-stage reasoning workflow, where heterogeneous event features are organized into reusable skills and orchestrated through a goal-oriented protocol. To facilitate evaluation, we construct a benchmark covering diverse query focuses. Experimental results show that TimelineAgent consistently outperforms existing methods, offering a practical framework for complex real-world timeline construction.
Qingkai Min, Yunxin Bao, Yifeng Lu +4
Zhejiang University · School of Engineering, Westlake University · Institute of Advanced Technology, Westlake Institute for Advanced Study +3
Existing LLM-based approaches to Architecture Decision Record (ADR) generation share a critical and largely unexamined assumption: that input is already reasonably structured. In practice, architectural decisions emerge from informal, noisy meetings where choices are implicit, fragmented, and entangled with off-topic dialogue, precisely the conditions under which single-pass prompting degrades. This paper presents GADR, a multi-agent, self-correcting workflow that extracts architectural decisions from raw meeting transcriptions and generates Nygard-formatted ADR drafts. A feasibility study comprising five real project meeting transcripts, expert review by four senior architects, and evaluation by fifteen students provides initial evidence that the agentic workflow captures most expert-identified decisions and produces drafts participants found clear and useful, outperforming zero-shot and few-shot baselines in stability and structural adherence. The study also addresses the underexplored trade-off of RAG-based enrichment improving ADR depth while simultaneously risking transcript-unfaithful content, raising open questions about traceability in automated architectural documentation that we believe is worth the community's attention.
Lucas Daniel Costa da Silva, Kiev Gama
Universidade Federal de Pernambuco (UFPE) Recife, Brasil