Build adaptive, cost-aware Graph-RAG pipelines that route queries through escalating retrieval stages (local -> bridge -> global) with triple-check verification and provenance map-back. Use when: 'build a graph RAG pipeline', 'implement adaptive retrieval for knowledge graphs', 'cost-aware multi-hop question answering', 'add evidence verification to RAG', 'handle mixed-difficulty queries efficiently', 'graph retrieval with source text grounding'.
Adapt general-purpose BPE tokenizers into domain- or language-specialized tokenizers using the AdaptBPE post-training strategy. Replaces low-utility tokens with high-frequency domain-specific tokens to improve tokenization efficiency without retraining from scratch. Trigger phrases: "adapt tokenizer to domain", "specialize BPE for medical text", "optimize tokenizer for French", "reduce token fertility for code", "adapt vocabulary for legal documents", "domain-specific tokenizer"
Explain generative AI outputs using the gSMILE perturbation-based attribution framework. Builds local surrogate models from controlled input perturbations and Wasserstein distance to produce token-level or word-level importance scores for LLM and diffusion model outputs. Triggers: 'explain why the model generated this', 'token attribution for prompt', 'which words in my prompt matter most', 'interpret generative model output', 'build explainability for my LLM pipeline', 'debug prompt influence on generation'
Automatically evaluate software research artifacts (code repositories with READMEs) by constructing dependency-aware command graphs, building containerized environments, and executing instructions with structured error recovery. Use when asked to: 'evaluate this artifact', 'reproduce this paper's results', 'run this repo's README instructions', 'check if this artifact builds and runs', 'automate artifact evaluation', 'verify research reproducibility'.
Design and implement OS-level resource controls for sandboxed AI agents using hierarchical cgroups, eBPF enforcement, and tool-call-level resource management. Use when: 'set up cgroups for AI agent containers', 'control memory for coding agents', 'isolate tool-call resources with eBPF', 'manage multi-tenant agent resource limits', 'prevent OOM kills in agent sandboxes', 'configure agent resource policies with cgroup v2'.
Build LLM-based multi-agent systems for supply chain inventory management using structured decision prompts and memory-retrieval (AIM-RM). Implements the beer game multi-echelon supply chain simulation with per-stage agents that use stepwise ordering prompts, safety-stock calculations, and Euclidean-distance memory retrieval of similar historical episodes. Use when asked to: "build a supply chain agent", "implement inventory management with LLMs", "create a beer game simulation with AI agents", "multi-agent ordering system", "AIM-RM memory retrieval agent", "supply chain decision prompt design".
Build intelligent alert lifecycle management systems for cloud infrastructure using graph-based denoising, RAG-powered summarization, and multi-agent rule refinement. Trigger phrases: - "reduce alert fatigue in our monitoring system" - "deduplicate and correlate alerts" - "summarize alerts for on-call engineers" - "refine our alerting rules automatically" - "build an alert denoising pipeline" - "too many alerts, help me triage"
Build multi-agent adaptive learning systems that diagnose knowledge gaps and recommend targeted resources. Implements the ALIGNAgent framework: Skill Gap Agent (proficiency estimation + concept-level diagnostic reasoning) and Recommender Agent (preference-aware resource retrieval aligned to deficiencies). Trigger phrases: - "Build an adaptive learning system" - "Create a personalized tutoring agent" - "Diagnose student knowledge gaps from quiz data" - "Build a skill gap analyzer for learners" - "Create an educational recommender that adapts to student performance" - "Implement a multi-agent pipeline for personalized education"