LADs is a six-agent framework for configuration, deployment, monitoring, and automated repair. Across 180 Dask, Redis, and Ray tasks, it achieves 68% deployment success and repairs 94% of runtime failures within three iterations.
@inproceedings{khan2025lads,title={LADs: Leveraging LLMs for AI-Driven DevOps},author={Khan, Saman Bahzad and others},booktitle={Conference on Machine Learning and Systems (MLSys)},year={2027},note={Under Submission. First author.},}
SkillGraph is a GraphRAG-based skill graph for extracting and reusing diagnostic reasoning from agent trajectories. The project builds a database of reasoning traces for skill extraction and retrieval on new system incidents, and uses a process-level critic that evaluates reasoning quality rather than final outcomes alone.
@inproceedings{khan2026skillgraph,title={SkillGraph: Reusable Reasoning Skills from LLM Agent Traces},author={Khan, Saman Bahzad and others},booktitle={Conference on Machine Learning and Systems (MLSys)},year={2027},note={Under Submission. First author.},}
An analysis of 535 queries across seven models finds reasoning traces ranging from under 1 KB to over 71 KB per query. The project develops co-location and indexing strategies for highly variable semantic reasoning workloads.
@inproceedings{khan2026agentmemory,title={Agent Memory: Storage and Retrieval of LLM Reasoning Traces at Scale},author={Khan, Saman Bahzad and others},booktitle={Conference on Machine Learning and Systems (MLSys)},year={2027},note={Under Submission},}
A federated framework in which clients share reusable skill templates and generalized reasoning patterns through a global indexed library while keeping private context and final adaptation local.
@inproceedings{khan2026federated,title={Federated Skill Sharing for LLM Agents},author={Khan, Saman Bahzad and others},booktitle={Conference on Machine Learning and Systems (MLSys)},year={2027},note={Under Submission},}
Sem-DPO is a weighted Direct Preference Optimization approach that reduces semantic drift during automated prompt optimization and establishes a theoretical bound on drift in preference-tuned prompt generators. Across three benchmarks and three language-model backbones, it improves CLIP similarity by 8–12% and human-preference scores by 5–9% over standard DPO, and is preferred in 82.2% of pairwise human evaluations.
@inproceedings{khan2025semdpo,title={Sem-DPO: Mitigating Semantic Inconsistency in Preference Optimization for Prompt Engineering},author={Mohamed, Anas and Khan, Azal Ahmad and Wang, Xinran and Khan, Ahmad Faraz and Ge, Shuwen and Khan, Saman Bahzad and Ahmad, Ayaan and Anwar, Ali},booktitle={Findings of the Association for Computational Linguistics: ACL 2026},year={2026},pages={23656--23674},doi={10.18653/v1/2026.findings-acl.1184},url={https://aclanthology.org/2026.findings-acl.1184/},}