Bhanu Prakash Vangala

Publications

My work spans LLM serving, reproducibility, trustworthy AI, and evaluation.

Google Scholar / Download CV

Published & accepted

  1. Per-language coverage and deferral rates for pooled and language-conditional thresholds

    One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification

    Bhanu Prakash Vangala, Navya Vangala

    Global South in AI Workshop at NeurIPS 2026 · Oral

    @misc{vangala2026thresholddoesfitlanguages,
      title={One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification},
      author={Bhanu Prakash Vangala and Vangala Navya},
      year={2026},
      eprint={2609.37861},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2609.37861},
}
  2. Clean-environment execution evaluation for AI-generated code

    AI-Generated Code Is Not Reproducible (Yet): An Empirical Study of Execution Reliability in LLM-Based Coding Agents

    Bhanu Prakash Vangala, Ashish Gehani, Tanu Malik

    ACM REP 2026 · Oral

    @inproceedings{vangala2026aigeneratedcodeexecution,
  author    = {Vangala, Bhanu Prakash and Gehani, Ashish and Malik, Tanu},
  title     = {AI-Generated Code Is Not Reproducible (Yet): An Empirical Study of Execution Reliability in LLM-Based Coding Agents},
  booktitle = {Proceedings of the 4th ACM Conference on Reproducibility and Replicability},
  series    = {ACM REP '26},
  publisher = {ACM},
  year      = {2026},
  month     = jul,
  pages     = {33--46},
  doi       = {10.1145/3820002.3828581},
  url       = {https://doi.org/10.1145/3820002.3828581}
}
  3. Pick and Spin routing and model orchestration architecture

    Pick and Spin: Cold-Start-Aware Routing for Self-Hosted LLM Serving

    Bhanu Prakash Vangala, Tanu Malik

    IEEE CLOUD 2026 · Oral

    @inproceedings{vangala2026pickspin,
  author    = {Vangala, Bhanu Prakash and Malik, Tanu},
  title     = {Pick and Spin: Cold-Start-Aware Routing for Self-Hosted LLM Serving},
  booktitle = {2026 IEEE 19th International Conference on Cloud Computing (CLOUD)},
  publisher = {IEEE},
  year      = {2026},
  month     = jul,
  pages     = {388--394},
  doi       = {10.1109/cloud72782.2026.00050},
  url       = {https://doi.org/10.1109/cloud72782.2026.00050}
}
  4. Measured eBPF explicit-read and memory-mapped consumption across scientific workflows

    CAMP: Consumption-Aware Memory Prediction for Scientific Workflows

    Nagarjuna Chowdary Kandimalla*, Bhanu Prakash Vangala*, Tanu Malik

    WORKS at SC26, the International Conference for High Performance Computing

    * Equal contribution

  5. Clean-environment execution evaluation for AI-generated code

    AI-Generated Code Is Not Reproducible (Yet): An Empirical Study of Dependency Gaps in LLM-Based Coding Agents

    Bhanu Prakash Vangala, Ali Adibifar, Ashish Gehani, Tanu Malik

    RAI Workshop at AAAI 2026 · Oral

    @misc{vangala2026aigeneratedcodereproducibleyet,
      title={AI-Generated Code Is Not Reproducible (Yet): An Empirical Study of Dependency Gaps in LLM-Based Coding Agents},
      author={Bhanu Prakash Vangala and Ali Adibifar and Ashish Gehani and Tanu Malik},
      year={2026},
      eprint={2512.22387},
      archivePrefix={arXiv},
      primaryClass={cs.SE},
      url={https://arxiv.org/abs/2512.22387},
}
  6. Multi-model serving and orchestration framework

    Efficient Multi-Model Orchestration for Self-Hosted Large Language Models

    Bhanu Prakash Vangala, Tanu Malik

    DAI Workshop at AAAI 2026 · Oral

    @misc{vangala2025efficientmultimodelorchestrationselfhosted,
      title={Efficient Multi-Model Orchestration for Self-Hosted Large Language Models},
      author={Bhanu Prakash Vangala and Tanu Malik},
      year={2025},
      eprint={2512.22402},
      archivePrefix={arXiv},
      primaryClass={cs.DC},
      url={https://arxiv.org/abs/2512.22402},
}
  7. HalluMat knowledge graph used for hallucination analysis

    HalluMat: Detecting Hallucinations in LLM-Generated Materials Science Content Through Multi-Stage Verification

    Bhanu Prakash Vangala, Sajid Mahmud, Pawan Neupane, Joel Selvaraj, Jianlin Cheng

    AAAI Spring Series 2025, AI for Engineering and Scientific Discoveries · Oral

    View the benchmark

    @misc{vangala2025hallumatdetectinghallucinationsllmgenerated,
      title={HalluMat: Detecting Hallucinations in LLM-Generated Materials Science Content Through Multi-Stage Verification},
      author={Bhanu Prakash Vangala and Sajid Mahmud and Pawan Neupane and Joel Selvaraj and Jianlin Cheng},
      year={2025},
      eprint={2512.22396},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2512.22396},
}
  8. HalluFormer transformer classification architecture

    HalluFormer: A Transformer-Based Framework for Detecting Hallucination in Large Language Models

    Sajid Mahmud*, Pawan Neupane*, Joel Selvaraj*, Bhanu Prakash Vangala*, Jianlin Cheng

    AAAI Spring Series 2025, AI for Engineering and Scientific Discoveries · Oral

    * Equal contribution

  9. A black-and-white portrait beside its colorized version

    Image Colorization using AI

    Bhanu Prakash Vangala, Abdul Mannan Khan, Sagar Sujith Somepalli, Rakesh Chigurupati, Pratham Shah, Shubham Nandlal Vishwakarma

    IJARESM 2022

Submitted & under review

Submitted manuscripts; these have not yet been accepted.

  1. Declared, installed, and traced software dependency sets

    Code That Works, Environments That Don’t: Measuring Environment Reproducibility in AI-Generated Software

    Bhanu Prakash Vangala, Tanu Malik

    AI Magazine — Special Issue on AI and Reproducibility (AAAI · Wiley) Under review

    Next version in progress: a public benchmark that tests how reliable code from AI coding agents really is, covering reproducibility, environment instability and dependency security. View the benchmark

    @misc{vangala2026codeworksenvironmentsdont,
      title={Code That Works, Environments That Don't: Measuring Environment Reproducibility in AI-Generated Software},
      author={Bhanu Prakash Vangala and Tanu Malik},
      year={2026},
      eprint={2610.00425},
      archivePrefix={arXiv},
      primaryClass={cs.SE},
      url={https://arxiv.org/abs/2610.00425},
}
  2. MasakhaNEWS learning curves across annotation budgets

    How Many Labels Does a Language Need? Annotation Budgets and Cross-Lingual Pooling for African-Language Text Classification

    Bhanu Prakash Vangala, Sowmya Guda, Navya Vangala

    Africa in AI Affinity Workshop at NeurIPS 2026 Under review

    @misc{vangala2026labelsdoeslanguageneed,
      title={How Many Labels Does a Language Need? Annotation Budgets and Cross-Lingual Pooling for African-Language Text Classification},
      author={Bhanu Prakash Vangala and Sowmya Guda and Navya Vangala},
      year={2026},
      eprint={2609.37882},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2609.37882},
}
  3. Chest X-ray classification research

    Evaluation Choices Shape Biomedical ML Claims: A Pediatric Pneumonia Benchmark Case Study

    Bhanu Prakash Vangala, Sowmya Guda, Latha Peddi, Navya Vangala

    RCMLR Workshop at NeurIPS 2026 Under review

    @misc{vangala2026evaluationchoicesshapebiomedical,
      title={Evaluation Choices Shape Biomedical ML Claims: A Pediatric Pneumonia Benchmark Case Study},
      author={Bhanu Prakash Vangala and Sowmya Guda and Latha Peddi and Navya Vangala},
      year={2026},
      eprint={2609.37848},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2609.37848},
}
  4. Attribution maps for held-out soybean images

    AI Sees, XAI Explains? Evaluating Explanation Reliability in Automated Seed Quality Inspection

    Bhanu Prakash Vangala, Navya Vangala

    Computers and Electronics in Agriculture (Elsevier · ScienceDirect) Under review

    @misc{vangala2026aiseesxaiexplains,
  author       = {Vangala, Bhanu Prakash and Vangala, Navya},
  title        = {AI Sees, XAI Explains? Evaluating Explanation Reliability in Automated Seed Quality Inspection},
  year         = {2026},
  howpublished = {SSRN},
  note         = {Available at SSRN: https://ssrn.com/abstract=7544801},
  doi          = {10.2139/ssrn.7544801},
  url          = {https://doi.org/10.2139/ssrn.7544801}
}
  5. Real soybean seeds compared with generated samples

    Visual Realism Does Not Predict Seed-Quality Utility in AI-Generated Synthetic Seeds

    Bhanu Prakash Vangala, Navya Vangala

    Artificial Intelligence in Agriculture (KeAi / Elsevier · ScienceDirect) Under review

  6. Real and generated maize kernels used in the cross-species check

    AI-Generated Seed Images Do Not Outperform Matched Reuse of Scarce Real Defect Images in Seed Quality Assessment

    Bhanu Prakash Vangala, Navya Vangala

    Scientific Reports (Nature Portfolio · Springer Nature) Under review

  7. Brain MRI classification research

    Scores That Hold, Benchmarks That Leak: Measuring Dataset Contamination in Public Brain-Tumor MRI Classification

    Bhanu Prakash Vangala, Sowmya Guda, Latha Peddi, Navya Vangala

    IEEE Journal of Biomedical and Health Informatics (IEEE · IEEE Xplore) Under review

    @misc{vangala2026scoresholdbenchmarksleak,
      title={Scores That Hold, Benchmarks That Leak: Measuring Dataset Contamination in Public Brain-Tumor MRI Classification},
      author={Bhanu Prakash Vangala and Sowmya Guda and Latha Peddi and Navya Vangala},
      year={2026},
      eprint={2610.00421},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2610.00421},
}

In preparation

  1. AdaptFlow plan enumeration, online optimization, and query execution

    AdaptFlow: Efficient Adaptation of Accuracy-First AI Workflows under Data Drift

    Bhanu Prakash Vangala, Shankar Aditya, Todd Neif, Abhilash Jindal, Tanu Malik

    Intended venue: MLSys 2027 In preparation

Posters & presentations

  1. Kubernetes LLM-as-a-Service deployment architecture

    Adaptive Inference: Orchestrating Fine-Tuned LLMs with Serverless GPUs in HPC Environments

    Bhanu Prakash Vangala, Tanu Malik

    University of Missouri · Poster

Earlier publications, preprints & theses