Trustworthy Agentic AI: Neuro-Symbolic Verification and Self-Correcting Reasoning in Multi-Agent Systems

About Internship

Dean Reserch and Development from AITekNix Center of Excellence, is accepting candidates interested in the field of Computer Science Engineering, in the research area of Artificial intelligence and Machine learning.

This research investigates verifiable, trustworthy agentic AI systems capable of high-level reasoning, multi-agent collaboration, and deterministic tool use in critical domains. While Large Language Models demonstrate strong generative capabilities, their stochastic nature introduces hallucinations, logical inconsistencies, and unpredictability in multi-step task execution. This project aims to bridge neural representations with formal symbolic logic and dynamic knowledge graphs. We will design self-correcting feedback loops, agentic planning heuristics, and constraint-guided verification mechanisms to ensure verifiable outputs. Interns will actively contribute to: 1. Literature Synthesis & Benchmarking: Curating evaluation datasets for multi-step reasoning, tool execution, and adversarial stress testing. 2. Architecture Design & Prototyping: Implementing agentic architectures using PyTorch, Hugging Face, LangGraph/AutoGen, and symbolic reasoners (e.g., Prolog, Z3, or Neo4j). 3. Empirical Evaluation: Measuring reasoning fidelity, hallucination reduction, latency, and computational efficiency against standard baselines. 4. Co-authorship: Drafting manuscripts targeted for submission to premier peer-reviewed conferences (e.g., NeurIPS, ICLR, AAAI) and releasing open-source benchmark suites