About

Sean Benhur

AI Engineer - Post Training - Bengaluru, India

I build LLM agents and the evaluation systems that tell you whether they actually work. Nearly five years across ML engineering and applied research: agent platforms, RAG pipelines, fine-tuning, post training, and eval frameworks.

Before industry, I published NLP research on low-resource Indian languages - Tamil in particular - and wrote about machine learning on Medium.

Currently open to ML/AI Opportunities.

PythonPyTorchLLM agents RAGEvaluationDatabricks RayKubernetesHugging Face

Experience

Senior AI Engineer - Navan · Jul 2026 – Present · Bengaluru

AI Core team, RnD Shared Services.

Background coding agent platform: engineered shared orchestration and context infrastructure around Cursor and Codex cloud runtimes - 200+ trigger-driven automations for bug triage and root-cause analysis across 10,000+ tickets; task turnaround down from 30 minutes to 5.

Shared agent memory platform: built a developer extension and memory pipeline on AWS Bedrock AgentCore, with repository-scoped knowledge sharing across developers and AI agents alongside isolated session memory; automated fact validation and memory pruning; supports 3,000+ developers.

Autonomous software modernization: built an autonomous coding agent that migrated a Java monolith feature into an independently deployable microservice - architecture and release plans, code changes, dependency resolution, tests and staging deploy - in 3 days versus an estimated month of manual work, with human steering.

Comcast - Chennai · Jan 2023 – Jun 2026

Machine Learning Engineer 3 (Apr 2025 – Jun 2026) and Machine Learning Engineer 2 (Jul 2023 – Apr 2025): agentic tool-calling architecture (MCP + A2A) for the Xfinity Assistant, cutting end-to-end latency 30% across 15M+ monthly users; automated evaluation framework for multi-agent systems - fine-tuned 7B user-simulation models, deterministic tool mocking, claim verification - validating workflows serving 20M+ monthly users. Earlier: natural-language-to-Elasticsearch platform, RAG log-intelligence pipeline (50% lower latency, 90% faster incident triage), GraphRAG vulnerability detection, and a multimodal support bot.

ML Engineering Intern (Jan – Jun 2023): model deployment and latency work for the Xfinity Assistant; built a RAG chatbot POC for technician queries. Comcast Spotlight Champ Award; first off-campus intern hire at Comcast India.

Data Science Consultant Intern - Tata Communications · Aug – Oct 2022 · Remote

Geospatial clustering features for a network expansion and planning application; Streamlit performance work.

Applied Research Intern, Low-Resource NLP - NVIDIA · Dec 2021 – Mar 2022 · Bengaluru

Built ~100 GB of monolingual corpora and pretrained GPT-style models for low-resource languages with Megatron-LM.

Research Intern - University of Galway · Nov 2021 – Apr 2022 · Ireland (remote)

Adapter-based transformers for offensive-language detection and multimodal misogyny detection in low-resource and code-mixed languages; work published at SemEval and in Computer Speech and Language.

Education

MSc Software Systems (5-year Integrated) - PSG College of Arts and Science, Coimbatore

Publications

S. Benhur, "Natural Language to Elasticsearch Query DSL," Computing Conference 2025, London.
C. Vasantharajan, S. Benhur, P. K. Kumarasen, R. Ponnusamy, S. Thangasamy, R. Priyadharshini, T. Durairaj, K. Sivanraju, A. Sampath, B. R. Chakravarthi, and J. P. McCrae, "TamilEmo: Finegrained Emotion Detection Dataset for Tamil," arXiv:2202.04725, 2022.
Ramaneswaran S, S. Benhur, and S. Ghosh, "Span Extraction Aided Improved Code-mixed Sentiment Classification," Proceedings of the Eighth Workshop on Noisy User-generated Text (W-NUT), 2022.
S. Mahadevan, S. Benhur, R. Nayak, M. Subramanian, K. Shanmugavadivel, K. Sivanraju, and B. R. Chakravarthi, "Transformers at SemEval-2022 Task 5: A Feature Extraction based Approach for Misogynous Meme Detection," Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval, at NAACL), 2022.
V. Palanikumar, S. Benhur, A. Hande, and B. R. Chakravarthi, "DE-ABUSE@TamilNLP-ACL 2022: Transliteration as Data Augmentation for Abuse Detection in Tamil," Proceedings of the Second Workshop on Speech and Language Technologies for Dravidian Languages (at ACL), 2022.
A. Sampath, T. Durairaj, B. R. Chakravarthi, R. Priyadharshini, S. C. Navaneethakrishnan, K. Shanmugavadivel, S. Thavareesan, S. Thangasamy, P. Krishnamurthy, A. Hande, S. Benhur, K. Ponnusamy, and S. Pandiyan, "Findings of the Shared Task on Emotion Analysis in Tamil," Proceedings of the Second Workshop on Speech and Language Technologies for Dravidian Languages (at ACL), 2022.
S. Benhur, R. Nayak, K. Sivanraju, A. Hande, S. C. Navaneethakrishnan, R. Priyadharshini, and B. R. Chakravarthi, "Hypers at ComMA@ICON: Modelling Aggressive, Gender Bias and Communal Bias Identification," Proceedings of the 18th International Conference on Natural Language Processing (ICON), ComMA Shared Task, 2021.
S. Benhur and K. Sivanraju, "Pretrained Transformers for Offensive Language Identification in Tanglish," Dravidian-CodeMix-HASOC 2021 (FIRE), CEUR Workshop Proceedings Vol. 3159.