Thoughtworks Research60https://research.thoughtworks.comA global technology consultancy that integrates strategy, design and engineering to drive digital innovationhttps://research.thoughtworks.com/etc.clientlibs/thoughtworks/clientlibs/clientlib-site/resources/images/favicon.svghttps://research.thoughtworks.com/etc.clientlibs/thoughtworks/clientlibs/clientlib-site/resources/images/favicon.svgResearchGeometric curriculum coverage for detecting summary incompleteness via Hausdorff distanceSat Jun 27 00:00:00 UTC 2026https://research.thoughtworks.com/library/geometric-curriculum-coverage-detecting-summary-incompletenessManikandan Ravikiran, Phillip Howard, Shayan Mohanty

Evaluating AI generated summaries is a critical challenge for educational AI systems. Explore Thoughtworks research into a technique that may be able to help.

]]>

Evaluating AI generated summaries is a critical challenge for educational AI systems. Explore Thoughtworks research into a technique that may be able to help.

]]>https://research.thoughtworks.com/library/geometric-curriculum-coverage-detecting-summary-incompletenessResearchHi-SEMFLOW: Lie algebra–based semantic flow for span-level informal language identification in HindiSat May 16 00:00:00 UTC 2026https://research.thoughtworks.com/library/hi-semflow-lie-algebra-based-semantic-flow-informal-language-identification-hindiManikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta, Rohit Saluja

Explore AI research into techniques that can effectively handle informal Hindi.

]]>

Explore AI research into techniques that can effectively handle informal Hindi.

]]>https://research.thoughtworks.com/library/hi-semflow-lie-algebra-based-semantic-flow-informal-language-identification-hindiResearchRethinking skeleton-based action recognition from an action-class prediction distribution perspectiveWed Apr 01 00:00:00 UTC 2026https://research.thoughtworks.com/library/rethinking-skeleton-based-action-recognition-action-class-prediction-distributionShuang Wu, Yingying Jiao, Haipeng Chen, Yingda Lyu, Yuheng Yang, Zhenguang Liu

Explore research into the long-standing computer vision challenge of action recognition.

]]>

Explore research into the long-standing computer vision challenge of action recognition.

]]>https://research.thoughtworks.com/library/rethinking-skeleton-based-action-recognition-action-class-prediction-distributionResearchDo tokenizers fail on informal Hindi expressions? Evidence from static, downstream and robustness analysesSun Mar 29 00:00:00 UTC 2026https://research.thoughtworks.com/library/do-tokenizers-fail-informal-hindi-expressionsManikandan Ravikiran, Shayan Mohanty, Tanmay Tiwari, Vibhu Gupta, Rakesh Prakash, Rohit Saluja

Explore AI research on the effectiveness of tokenizers on informal Hindi expressions.

]]>

Explore AI research on the effectiveness of tokenizers on informal Hindi expressions.

]]>https://research.thoughtworks.com/library/do-tokenizers-fail-informal-hindi-expressionsResearchTinySQL: A progressive text-to-SQL dataset for mechanistic interpretability research Tue Mar 17 00:00:00 UTC 2026https://research.thoughtworks.com/library/tinysql-progressive-text-to-sql-datasetAmirali Abdullah, Abir Harrasse, Philip Quirke, Clement Neo, Dhruv Nathawani, Luke MarksTinySQL: A progressive text-to-SQL dataset for mechanistic interpretability research

]]>TinySQL: A progressive text-to-SQL dataset for mechanistic interpretability research

]]>https://research.thoughtworks.com/library/tinysql-progressive-text-to-sql-datasetResearchA semantic parsing framework for end-to-end time normalizationSat Dec 20 00:00:00 UTC 2025https://research.thoughtworks.com/library/semantic-parsing-framework-end-to-end-time-normalizatoinXin Su, Phillip Howard, Sungduk Yu and Steven Bethard

Explore AI research into a new framework for time normalization.

]]>

Explore AI research into a new framework for time normalization.

]]>https://research.thoughtworks.com/library/semantic-parsing-framework-end-to-end-time-normalizatoinResearchCultural awareness in vision-language models: A cross-country explorationMon Dec 15 00:00:00 UTC 2025https://research.thoughtworks.com/library/cultural-awareness-vision-language-modelsPhillip Howard, Avinash Madasu, Vasudev Lal

AI researchers explore how vision-language models perpetuate national stereotypes and assumptions.

]]>

AI researchers explore how vision-language models perpetuate national stereotypes and assumptions.

]]>https://research.thoughtworks.com/library/cultural-awareness-vision-language-modelsResearchMultiFeature graph convolutional network for OCR verificationWed Dec 10 00:00:00 UTC 2025https://research.thoughtworks.com/library/multifeature-graph-convolutional-network-ocrManikandan Ravikiran, Shikhar Dubey, Krish Mittal, Sourava Kumar Behera, Nitin Kumar, Saurabh Shigwan, Rohit SalujaMultiFeature graph convolutional network for OCR verification

]]>MultiFeature graph convolutional network for OCR verification

]]>https://research.thoughtworks.com/library/multifeature-graph-convolutional-network-ocrResearchRetrieval augmented forecasting for generalizationMon Dec 08 00:00:00 UTC 2025https://research.thoughtworks.com/library/retrieval-augmented-forecasting-generalizationManikandan Ravikiran, Aditi Gautam, Fatma Sena EkizRetrieval augmented forecasting for generalization

]]>Retrieval augmented forecasting for generalization

]]>https://research.thoughtworks.com/library/retrieval-augmented-forecasting-generalizationResearchBeyond MAE: Measuring forecast reliability with temporal dependence-aware error (TDE)Mon Dec 08 00:00:00 UTC 2025https://research.thoughtworks.com/library/beyond-mae-measuring-forecast-reliability-tdeManikandan Ravikiran, Aditi Gautam, Alisha ChulaniBeyond MAE: Measuring forecast reliability with temporal dependence-aware error (TDE)

]]>Beyond MAE: Measuring forecast reliability with temporal dependence-aware error (TDE)

]]>https://research.thoughtworks.com/library/beyond-mae-measuring-forecast-reliability-tdeResearchA superpersuasive autonomous policy debating systemSat Nov 22 00:00:00 UTC 2025https://research.thoughtworks.com/library/superpersuasive-autonomous-policy-debating-systemAllen Roush, Devin Gonier, John Hines, Judah Goldfeder, Philippe Martin Wyder, Sanjay Basu, Ravid Shwartz Ziv

Find out how a team of AI researchers, including one Thoughtworker, developed an autonomous policy debating system.

]]>

Find out how a team of AI researchers, including one Thoughtworker, developed an autonomous policy debating system.

]]>https://research.thoughtworks.com/library/superpersuasive-autonomous-policy-debating-systemResearchPruning the paradox: How CLIP’s most informative heads enhance performance while amplifying biasThu Nov 20 00:00:00 UTC 2025https://research.thoughtworks.com/library/pruning-paradox-clip-informative-headsPhillip Howard, Avinash Madasu, Vasudev LalPruning the paradox: How CLIP’s most informative heads enhance performance while amplifying bias

]]>Pruning the paradox: How CLIP’s most informative heads enhance performance while amplifying bias

]]>https://research.thoughtworks.com/library/pruning-paradox-clip-informative-headsResearchTransformer-based temporal information extraction and application: A reviewTue Nov 04 00:00:00 UTC 2025https://research.thoughtworks.com/library/transformer-based-temporal-information-extraction-application-reviewXin Su, Phillip Howard, Steven Bethard

Two Thoughtworkers contributed to a research paper on transformer-based temporal information extraction and application. Read it now.

]]>

Two Thoughtworkers contributed to a research paper on transformer-based temporal information extraction and application. Read it now.

]]>https://research.thoughtworks.com/library/transformer-based-temporal-information-extraction-application-reviewResearchAntislop: A comprehensive framework for identifying and eliminating repetitive patterns in language modelsThu Oct 16 00:00:00 UTC 2025https://research.thoughtworks.com/library/antislop-framework-identifying-eliminating-repetitive-patterns-language-modelsAllen Roush, Samuel Paech, Judah Goldfeder, Ravid Shwartz-Ziv

Explore AI research that attempts to minimize repetitive patterns in AI-generated language.

]]>

Explore AI research that attempts to minimize repetitive patterns in AI-generated language.

]]>https://research.thoughtworks.com/library/antislop-framework-identifying-eliminating-repetitive-patterns-language-modelsResearchp-less Sampling: A robust hyperparameter-free approach for LLM decodingSat Sep 27 00:00:00 UTC 2025https://research.thoughtworks.com/library/p-less-sampling-robust-hyperparameter-free-approach-llm-decodingRunyan Tan, Phillip Howard, Shuang Wu

Explore a new technique for LLM decoding developed by AI researchers at Thoughtworks.

]]>

Explore a new technique for LLM decoding developed by AI researchers at Thoughtworks.

]]>https://research.thoughtworks.com/library/p-less-sampling-robust-hyperparameter-free-approach-llm-decodingResearchBeyond I am sorry, I can’t: dissecting large language model refusalSun Sep 07 00:00:00 UTC 2025https://research.thoughtworks.com/library/beyond-i-am-sorry-i-can-t-dissecting-large-language-model-refusalAmirali Abdullah, Nirmalendu Prakash, Yeo Wei Jie, Ranjan Satapathy, Erik Cambria, Roy Ka Wei Lee

Explore how large language models generate refusals, why they happen and what they reveal about AI safety, alignment and model behavior.

]]>

Explore how large language models generate refusals, why they happen and what they reveal about AI safety, alignment and model behavior.

]]>https://research.thoughtworks.com/library/beyond-i-am-sorry-i-can-t-dissecting-large-language-model-refusalResearchTowards transparent AI grading: Entropy as a signal for human-AI disagreementWed Aug 06 00:00:00 UTC 2025https://research.thoughtworks.com/library/transparent-ai-grading-entropy-signalManikandan Ravikiran, Karrtik Iyer, Shayan Mohanty, Prasanna PendseTowards transparent AI grading: Entropy as a signal for human-AI disagreement

]]>Towards transparent AI grading: Entropy as a signal for human-AI disagreement

]]>https://research.thoughtworks.com/library/transparent-ai-grading-entropy-signalResearchInvestigating the robustness of retrieval-augmented generation at the query levelSun Jul 20 00:00:00 UTC 2025https://research.thoughtworks.com/library/robustness-retrieval-augmented-generationPhillip Howard, Xin Su, Sezen Perçin, Qutub Sha Syed, Aleksei Kuvshinov, Leo Schwinn, Kay-Ulrich SchollInvestigating the robustness of retrieval-augmented generation at the query level

]]>Investigating the robustness of retrieval-augmented generation at the query level

]]>https://research.thoughtworks.com/library/robustness-retrieval-augmented-generationResearchLie Algebra based semantic flow for incompleteness detection in summarizationFri Jun 20 00:00:00 UTC 2025https://research.thoughtworks.com/library/lie-algebra-semantic-flow-incompletenessManikandan Ravikiran, Shayan Mohanty, Veeraraju Elluru, Karrtik Iyer, Prasanna PendseLie Algebra based semantic flow for incompleteness detection in summarization

]]>Lie Algebra based semantic flow for incompleteness detection in summarization

]]>https://research.thoughtworks.com/library/lie-algebra-semantic-flow-incompletenessResearchBeyond linear steering: Unified multi-attribute control for language modelsFri May 30 00:00:00 UTC 2025https://research.thoughtworks.com/library/beyond-linear-steering-multi-attribute-controlAmirali Abdullah, Narmeen Oozeer, Luke Marks, Shreyans Jain, Fazl Barez

K-steering is a technique designed for controlling multiple behavioral attributes in LLMs at inference. Learn more about it and how researchers developed it.

]]>

K-steering is a technique designed for controlling multiple behavioral attributes in LLMs at inference. Learn more about it and how researchers developed it.

]]>https://research.thoughtworks.com/library/beyond-linear-steering-multi-attribute-controlResearchTraining-free mitigation of language reasoning degradation after multimodal instruction tuningWed May 28 00:00:00 UTC 2025https://research.thoughtworks.com/library/training-free-mitigation-language-reasoning-degradationPhillip Howard, Xin Su, Neale Ratzlaff, Man Luo, Vasudev LalTraining-free mitigation of language reasoning degradation after multimodal instruction tuning

]]>Training-free mitigation of language reasoning degradation after multimodal instruction tuning

]]>https://research.thoughtworks.com/library/training-free-mitigation-language-reasoning-degradationResearchLearning from reasoning failures via synthetic data generationSun Apr 20 00:00:00 UTC 2025https://research.thoughtworks.com/library/learning-reasoning-failures-synthetic-data-generationPhillip Howard, Gabriela Ben Melech Stan, Estelle Aflalo, Avinash Madasu, Vasudev Lal

What can the reasoning failures of a large multimodal model teach us about synthetic data generation? Learn more in this AI research paper.

]]>

What can the reasoning failures of a large multimodal model teach us about synthetic data generation? Learn more in this AI research paper.

]]>https://research.thoughtworks.com/library/learning-reasoning-failures-synthetic-data-generationResearchIs your paper being reviewed by an LLM? Benchmarking AI text detection in peer reviewWed Feb 26 00:00:00 UTC 2025https://research.thoughtworks.com/library/is-paper-reviewed-llm-benchmarking-ai-text-detectionPhillip Howard, Sungduk Yu, Man Luo, Avinash Madasu, Vasudev Lal

Explore research seeking to benchmark AI text detection in academic peer review.

]]>

Explore research seeking to benchmark AI text detection in academic peer review.

]]>https://research.thoughtworks.com/library/is-paper-reviewed-llm-benchmarking-ai-text-detectionResearchScaling knowledge graph construction through synthetic data generation and distillationTue Oct 22 00:00:00 UTC 2024https://research.thoughtworks.com/library/scaling-knowledge-graph-construction-synthetic-data-generation-distillationXin Su, Phillip Howard, Prafulla Kumar Choubey, Man Luo, Xiangyu Peng, Caiming Xiong, Tiep Le, Shachar Rosenman, Vasudev Lal, Phil Mui, Ricky Ho, Chien-Sheng Wu

How can synthetic data and distillation help teams construct larger knowledge graphs? This research highlights techniques worth exploring.

]]>

How can synthetic data and distillation help teams construct larger knowledge graphs? This research highlights techniques worth exploring.

]]>https://research.thoughtworks.com/library/scaling-knowledge-graph-construction-synthetic-data-generation-distillation