{ "generated": "2026-08-20T20:10:21.654Z", "source": "https://profile.vballoli.com/agents/", "person": { "name": "Vaibhav Balloli", "role": "Applied AI Researcher and Computer Science Ph.D. Candidate", "affiliation": "University of Michigan", "advisor": "Elizabeth Bondi-Kelly", "email": "balloli.vb@gmail.com", "url": "https://profile.vballoli.com/", "researchAreas": [ "Large Language Models", "Reinforcement Learning", "Representation Learning", "Human-AI Interaction", "Agentic Systems", "Machine Learning" ], "summary": "Applied AI researcher working on the reliability, safety, and usability of AI systems built with and for people. My Ph.D. research focuses on improving conversational (human–LLM) and agentic capabilities for maternal health." }, "identifiers": { "orcid": "0000-0003-1036-9672", "dblp": "348/6962", "semanticScholar": "1564592237" }, "profiles": [ "https://github.com/vballoli", "https://scholar.google.com/citations?user=-6i7e7sAAAAJ&hl=en", "https://orcid.org/0000-0003-1036-9672", "https://www.semanticscholar.org/author/1564592237", "https://dblp.org/pid/348/6962", "https://www.linkedin.com/in/vaibhavballoli", "https://twitter.com/v_balloli" ], "education": [ { "institution": "University of Michigan, Ann Arbor", "qualification": "Ph.D. Candidate, M.S. in Computer Science and Engineering", "period": "2023 – present" }, { "institution": "BITS Pilani, Hyderabad Campus", "qualification": "B.E. in Electronics and Communication Engineering", "period": "2016 – 2020" } ], "experience": [ { "organisation": "Microsoft Research Redmond, AI Frontiers", "role": "Research Intern", "period": "May – Aug 2025", "summary": "Multi-turn reinforcement learning with large language model agents, and agent evaluation. Advised by Hussein Mozannar, Adam Fourney, Gagan Bansal, Saleema Amershi, and Eric Horvitz." }, { "organisation": "Microsoft Research India", "role": "Research Fellow", "period": "Jun 2022 – Jun 2023", "summary": "Reinforcement learning and large language models on Project Vasudha and Project VeLLM. Advised by Dr. Akshay Nambi and Tanuja Ganu." }, { "organisation": "Microsoft Research India", "role": "Societal Impact through Cloud and AI Fellow", "period": "Jun 2021 – Jun 2022", "summary": "Computer vision, bandits, and localisation for HAMS. Advised by Dr. Akshay Nambi, Tanuja Ganu, and Dr. Venkat Padmanabhan." } ], "systems": [ { "name": "HAMS: automated driver licence testing", "status": "Deployed", "organisation": "Microsoft Research India", "period": "2021 – 2022", "contribution": "Societal Impact through Cloud and AI Fellow. Worked on the computer vision, bandit, and localisation components as the system scaled.", "outcome": "Smartphone-based automated licence testing deployed across India, serving 300,000+ candidates and replacing a subjective manual assessment with an auditable one.", "areas": [ "Computer vision", "On-device inference", "Bandits", "Localisation" ], "topics": [ "vision", "social-impact" ], "paper": null, "links": [ { "label": "Project", "url": "https://www.microsoft.com/en-us/research/project/hams/" }, { "label": "Deployment", "url": "https://www.microsoft.com/en-us/research/project/hams/automated-driver-license-testing/" }, { "label": "Press", "url": "https://news.microsoft.com/en-in/features/microsoft-ai-automates-drivers-license-test-india/" } ] }, { "name": "PATHFinder: prenatal care planning agent", "status": "In progress", "organisation": "University of Michigan", "period": "2025 – 2026", "contribution": "Lead author. Designed and built the four-stage agent: patient intake, dynamic interaction, plan synthesis, and clinician oversight.", "outcome": "Conversational agent that produces individualised prenatal care plans aligned to ACOG's PATH guidelines and surfaces local resources. Frontier models scored against expert rubrics across five clinical dimensions, with clinician evaluation in Michigan health systems underway.", "areas": [ "LLM agents", "Long-context optimisation", "Clinical rubrics" ], "topics": [ "llms", "healthcare", "evals" ], "paper": "https://profile.vballoli.com/papers/pathfinder/", "links": [ { "label": "Project", "url": "https://realize-lab.github.io/PATHFinder" }, { "label": "arXiv", "url": "https://arxiv.org/abs/2607.24768" } ] }, { "name": "CHAIR & SEEK-CBM: steerable wildlife re-identification", "status": "In progress", "organisation": "University of Michigan · MIT", "period": "2023 – 2026", "contribution": "Lead author on CHAIR; co-author on the SEEK-CBM follow-up with the Beery Lab at MIT.", "outcome": "Concept-bottleneck retrieval that lets conservation experts intervene mid-search, built for platforms like ElephantBook and evaluated on elephant populations in the Masai Mara. SEEK-CBM's modern backbones improved retrieval performance by 497% over CHAIR.", "areas": [ "Concept bottleneck models", "Self-supervised vision", "Human-in-the-loop retrieval" ], "topics": [ "vision", "social-impact" ], "paper": "https://profile.vballoli.com/papers/seek-cbm/", "links": [ { "label": "Project", "url": "https://realize-lab.github.io/CHAIR" }, { "label": "Code", "url": "https://github.com/realize-lab/CHAIR" }, { "label": "Talk", "url": "https://www.youtube.com/watch?v=Czo5YQU4-WY" } ] }, { "name": "VeLLM: multilingual LLMs for the global majority", "status": "Prototyping", "organisation": "Microsoft Research India", "period": "2022 – 2023", "contribution": "Research Fellow. Co-led the dynamic prompt/model/embedding selection work that became the COLING 2025 paper.", "outcome": "Platform work behind LLM applications piloted in schools across India, improving coverage for non-Latin-script and low-resource languages without additional training.", "areas": [ "LLMs", "Multilingual evaluation", "Prompt & model routing" ], "topics": [ "llms", "social-impact" ], "paper": "https://profile.vballoli.com/papers/leap/", "links": [ { "label": "Project", "url": "https://www.microsoft.com/en-us/research/project/project-vellm/" }, { "label": "Context", "url": "https://www.microsoft.com/en-us/research/story/advancing-ai-to-meet-needs-of-the-global-majority/" } ] } ], "topics": [ { "id": "social-impact", "label": "AI for Social Impact", "description": "Deployments where the measure of success is whether something changed for the people involved: road safety testing across India, elephant re-identification for conservation partners, prenatal care planning with clinicians in the loop. The recurring lesson is that the deployment constraints reshape the model, not the other way round.", "paperCount": 4, "url": "https://profile.vballoli.com/topics/social-impact/" }, { "id": "healthcare", "label": "AI for Health", "description": "Research on AI in peripartum and reproductive health, spanning the full arc from formative study to deployed system to benchmark: what birthing people, clinicians, doulas, and community health workers actually need from an information tool, an agent built to those requirements, and a dataset for checking whether the surrounding information ecosystem is accurate.", "paperCount": 3, "url": "https://profile.vballoli.com/topics/healthcare/" }, { "id": "llms", "label": "Large Language Models", "description": "Work on making large language models useful when the stakes are real: choosing prompts, models, and embeddings at run time rather than fixing them in advance, building conversational agents that gather context before they advise, and measuring where frontier models still fall short. The thread running through it is that an LLM is a component in a system, not the system itself.", "paperCount": 3, "url": "https://profile.vballoli.com/topics/llms/" }, { "id": "vision", "label": "Computer Vision", "description": "Vision systems designed to be steered by the people who use them. This includes concept-bottleneck retrieval for wildlife re-identification, where a domain expert can intervene on intermediate concepts mid-search, and adaptive streaming perception that trades accuracy against latency at run time based on scene content and system contention.", "paperCount": 3, "url": "https://profile.vballoli.com/topics/vision/" }, { "id": "evals", "label": "Evaluation & Benchmarks", "description": "Datasets and evaluation protocols built with the domain experts who have to trust the result. Rather than scoring models on proxies, this work curates expert-annotated ground truth (clinicians reviewing health claims, rubrics written by practising obstetricians) and asks whether language models can stand in for that judgement. Usually the answer is partly, and the interesting part is where the gap sits.", "paperCount": 2, "url": "https://profile.vballoli.com/topics/evals/" } ], "workingPapers": [ { "title": "To Ask or Not To Ask: Evaluating Information Seeking, Execution, and Abstention Capabilities in Long-Context LLM Agents", "authors": "Vaibhav Balloli, Elizabeth Bondi-Kelly" }, { "title": "Towards Improving Interactivity in Multi-Turn Human-Agent Systems", "authors": "Vaibhav Balloli, Hussein Mozannar, Adam Fourney, Gagan Bansal, Saleema Amershi, Eric Horvitz" } ], "service": [ { "label": "Conference reviewing", "detail": "NeurIPS 2026, KDD 2026, NeurIPS 2025, IJCAI 2025, ICML 2025, AAMAS 2025, ICLR 2025, NeurIPS 2024" }, { "label": "Workshop reviewing", "detail": "CV4Animals @ CVPR 2024, CompSust @ NeurIPS 2023, EvalEval @ ACL 2026" } ] }