Sushrita Rakshit

Ph.D. in Computer Science

B.S.E. in Computer Science, Minor in Statistics

Coursework: Machine Learning, Reinforcement Learning, Computer Vision, Web Systems, Databases, UX Design, Accessibility-Driven Design, Probability, Regression, Statistical Computing, Computational Social Science, and Computational Linguistics.

Bi-Align Lab, Remote / Shanghai, China

Paper(s): Pseudo-Deliberation in Language Models: When Reasoning Fails to Align Values and Actions, Position: Towards Bidirectional Human-AI Alignment

Investigated whether the values language models articulate in reasoning actually carry over into the dialogue they generate.

BlablabLAB, University of Michigan

Paper(s): Causally Modeling the Linguistic and Social Factors that Predict Email Response, Real or Robotic? Assessing Whether LLMs Accurately Simulate Qualities of Human Responses in Dialogue

Investigated how humans and large language models reason, communicate, and respond across socially grounded language tasks. Focused on human communication patterns and how language models can represent or simulate those patterns.

USC Affective Computing Group, Los Angeles, CA

Paper(s): Emotionally-Aware Agents for Dispute Resolution, KODIS: A Multicultural Dispute Resolution Dialogue Corpus

Investigated how humans express emotion in cross-cultural dispute resolution dialogues and how large language models can recognize, model, and respond to those emotional signals.

Advanced Propulsion Concepts Lab, Ann Arbor, MI

Investigating how course discussion data can be used to build retrieval-based language model tools that help students find relevant answers from prior Piazza interactions.

Verizon, Tampa, FL

• Built a technician-facing multi-agent assistant with RAG, dynamic API calls, and tool use to support real-time field troubleshooting.
• Engineered a CNN-based wake-word detection system trained on thousands of audio examples to launch and control an internal application through voice activation.
• Served as primary system architect for a troubleshooting wizard, designing the Neo4j graph and streaming structure used to model complex customer support workflows.
• Developed AI-assisted SQL generation for large-scale workforce planning, dynamically querying and hydrating tables to track technician schedules, workloads, and assignments.

• Verizon Spotlight Award, 2025.
• Computing Research Association Outstanding Undergraduate Research Award, Honorable Mention, 2025.
• NSF Research Experiences for Undergraduates, $8,650, 2024.
• University of Michigan Regional Merit Scholarship, $1,500, 2021.

• Emotionally Aware Agents for Dispute Resolution, ACII Oral Presentation, Canberra, Australia.
• Emotional Modeling in Disputes with Statistical Learning and Large Language Models, SoCal Summer Research Symposium 2024.
• Interpreting Spatial Reasoning Capabilities of Large Language Models, Midwestern Speech and Language Days 2024.

University of Michigan School of Information

• SI 301: Models of Social Information Processing — supported 130+ students through office hours, discussion sections, grading, and rubrics.
• SI 339: Web Design, Development, and Accessibility — co-taught discussions, led office hours, and supported coding exercises.

• Saptak Academy — coordinated free music workshops and supported student practice.
• 826 Michigan — tutored students ages 8–14 and supported nonprofit shop operations.
• Reviewed for the Bi-directional Human-AI Alignment Workshop at ICLR 2025.
• Reviewed for the Bi-directional Human-AI Alignment Workshop at CHI 2025.