Consensus collapse and evidence independence
Studying when apparent agreement among agents reflects shared or insufficient evidence rather than independent support.
Cincinnati, Ohio, United States
AI researcher · Computer Science PhD
Ritesh Janga
AI Research Assistant at P&G Digital Accelerator @ UC · PhD in Computer Science
My current research examines reliability in multi-agent LLM systems: when agreement masks missing evidence, how agents should respond to evidence changes, and how causal audits can distinguish collusion from correlated calibration.
- Current role
- AI Research Assistant
P&G Digital Accelerator @ UC - Education
- PhD in Computer Science
University of Cincinnati - Research
- Multi-Agent LLM Reliability
Evidence Sensitivity · Causal Auditing
Profile
About
I am an AI Research Assistant at the P&G Digital Accelerator at the University of Cincinnati and a PhD student in Computer Science.
My current research investigates failure modes and trustworthy decision-making in multi-agent LLM systems, including consensus collapse in retrieval-augmented generation, evidence-sensitive agent re-invocation, and causal auditing of apparent collusion.
I design methods that test whether agent agreement is independently supported, identify which agents should be re-executed when evidence changes, and separate coordinated behavior from shared calibration effects. My earlier work focused on federated learning for privacy-preserving agriculture, computer vision, and applied data systems.
Scholarship
Research
Questions, methods, and application domains shaping my work.
Selective agent re-invocation
Calibrating evidence-change scores to re-execute only the agents whose outputs are materially affected.
Collusion versus correlated calibration
Using causal-mediation analysis to distinguish coordinated behavior from correlations induced by shared calibration.
Current research
Conference submissionsRetrieval-Induced Consensus Collapse: Agreement Is Not Independent Evidence in Multi-Agent RAG
Ritesh Janga, Justin Zhan
ESS-DAG: Calibrated Evidence Sensitivity Scoring for Selective Agent Re-Invocation in Multi-Agent Pipelines
Ritesh Janga, Justin Zhan
Distinguishing Collusion from Correlated Calibration: A Causal-Mediation Audit Framework and Benchmark for LLM Multi-Agent Systems
Tiffany Zhan, Ritesh Janga, Justin Zhan
Publications
Complete Google Scholar profile ↗A Collaborative Intelligence Framework: Federated Learning for Privacy-Preserving Crop Disease Detection
Ritesh Janga, Rushit Dave, Rachna Shah, Mansi Bhavsar · ACDSA 2026
DOI ↗Impact of Cross-Client Heterogeneity in Federated Learning for Real-World Plant Disease Classification
Ritesh Janga · Master’s thesis, Minnesota State University, Mankato
Thesis ↗Enhancing Smart Farming Through Federated Learning: A Secure, Scalable, and Efficient Approach for AI-Driven Agriculture
Ritesh Janga, Rushit Dave · Artificial Intelligence and Applications
DOI ↗Harvesting Intelligence through Artificial Intelligence Applications for Precision and Productivity in Agriculture
Rachna Shah, Gaurav Jeet, Ritesh Janga · ICIDCA 2025
Advancing Dermatology with Deep Learning: An Ensemble Model for Classifying Skin Lesion
D. Chandra Mouli, R. Abhishek, J. Ritesh et al. · International Journal of Scientific Research & Engineering Trends
Paper ↗Portfolio
Selected projects
Research prototypes and applied systems from my public GitHub portfolio.
Proactive Email Agent
A four-level autonomy system combining hard safety rules, preference learning, risk certification, an evaluation harness, and tests.
- Python
- Agent safety
- Calibration
AgriFedNet
Privacy-preserving plant-disease classification across heterogeneous clients with Flower and MobileNetV3-Small.
- Flower
- PyTorch
- MobileNetV3
AI Doctor Chatbot
A LoRA-adapted medical-information chatbot with a Streamlit interface.
- Hugging Face
- PEFT
- LoRA
Wildfire Prediction Pipeline
Contributed to an environmental pipeline integrating streaming, distributed storage, modeling, and visualization.
- Kafka
- Spark
- Docker
Skin Lesion Classification
Contributed to a U-Net segmentation and ensemble-classification workflow for dermatological images.
- U-Net
- DenseNet
- ResNet
Human Presence Monitoring
A Python system for real-time human-presence detection, monitoring, and database-backed records.
- Python
- Computer vision
- Dashboard
Career
Experience
Jun 2026—Present
AI Research Assistant
P&G Digital Accelerator @ UC
Cincinnati, Ohio, United States
Jan 2025—May 2026
Graduate Assistant · Simulation Technologist
Minnesota State University, Mankato
Mankato, Minnesota, United States
Jan 2024—Jun 2024
AI/ML Intern
Fluentgrid Limited
Worked on vehicle number-plate identification using EasyOCR, YOLOv4, and custom-data training for text extraction.
Jun 2023—Jul 2023
Machine Learning Intern
Fluentgrid Limited
Analyzed power consumption and developed a person-counting system with a Python dashboard and database storage for timestamps, counts, and images.
Apr 2023—Jun 2023
Python Developer
ZENCHI SOFTWARES (OPC) PRIVATE LIMITED
Applied cookie scraping to support market insights and user-experience analysis.
Academic background
Education
Doctor of Philosophy · Computer Science
University of Cincinnati — College of Engineering and Applied Science
Master’s Degree · Data Science
Minnesota State University, Mankato
Bachelor of Technology · Artificial Intelligence and Data Science
Vignan’s Institute of Information Technology
Intermediate Education
Sri Chaitanya College of Education
Secondary Education
Dr. KKR’s Gowtham High School
Updates
Latest news
Released an updated Proactive Email Agent repository and design notes.
View update ↗Joined P&G Digital Accelerator @ UC as an AI Research Assistant.
Began the Computer Science PhD program at the University of Cincinnati.
Served as a peer reviewer for the IEEE Computational Intelligence Society.
Received a Graduate Student Research Grant.
Document ↗Published the ACDSA paper on privacy-preserving crop-disease detection.
DOI ↗Community
Volunteering & service
Research leadership & review
Lead Researcher
Research Den · Directing a voluntary research team through methodology design, data processing, and manuscript development.
Research Mentor
University of Cincinnati
Peer Reviewer · IEEE Transactions on Artificial Intelligence
IEEE Computational Intelligence Society
Peer Reviewer · Control and Cybernetics
IBS PAN — Systems Research Institute, Polish Academy of Sciences
Peer Reviewer · OzCHI
Australian Conference on Computer-Human Interaction
Mentoring & community
Peer Mentor
Minnesota State University, Mankato
Student Volunteer
Twin Cities ACM Chapter
Student Volunteer
Minnesota State University, Mankato
Technical Secretary
OWASP VIGNAN
Web App Design Coordinator
Vignan’s Institute of Information Technology · Vista2k22
Food Coordinator Lead
Vignan’s Institute of Information Technology
Internshala Student Partner 23
Internshala
Recognition
Honors & awards
Graduate Incentive Award
University of Cincinnati · Associated with P&G Digital Accelerator @ UC. Competitive graduate research assistantship with tuition remission and a financial stipend.
Graduate Student Research Grant
College of Graduate Studies and Research · Minnesota State University, Mankato. Awarded for the MS thesis project on privacy-preserving crop-disease detection.
View document ↗Graduate Student Scholarship
Minnesota State University, Mankato · Competitive graduate assistantship with tuition remission and a financial stipend.
Recommendation · May 28, 2026
“No shortcuts, no excuses, just genuine effort, day after day.”
Dr. Rushit Dave · Advisor and Director of Computer Science, Minnesota State University, Mankato
Dr. Dave highlights Ritesh’s ownership of his thesis, independence, initiative, consistency, work ethic, and character.
Beyond research
University Badminton
Gold medal in doubles and two bronze medals in singles · Minnesota State University, Mankato
Professional development
Licenses & certifications
NVIDIA DLIGetting Started with AI on Jetson NanoCredential ↗
DataproC ProgrammingCredential ↗
Grafx IT SolutionsPython with Machine LearningCredential ↗
MicrosoftCreate Dashboards in Power BICredential ↗
UdemyWeb Scraping in PythonCredential ↗
LinkedIn LearningUbuntu Linux: Essential CommandsCredential ↗
LinkedIn LearningGit Essential TrainingCredential ↗
LinkedIn LearningLearning Power BI DesktopCredential ↗
Cisco Networking AcademyProgramming Essentials in Python — PCAPCredential ↗
APSSDCMicrosoft OfficeCredential ↗
Expertise
Skills
Research areas
Reliable LLMs · Agentic AI · RAG · GraphRAG · Federated Learning · Privacy-Preserving ML
AI & modeling
PyTorch · TensorFlow · scikit-learn · Deep Learning · NLP · Computer Vision · LLM Fine-Tuning · PEFT · LoRA
Data systems
Python · SQL · R · ETL · Spark · Kafka · Data Warehousing · Streaming Data
Platforms & databases
Docker · Git · Streamlit · Power BI · MongoDB · PostgreSQL · SQL Server · MySQL · Oracle · Cassandra
Contact
Research, collaboration, and conversation.
I welcome opportunities to collaborate on federated learning, applied machine learning, and responsible intelligent systems.