{ "schema": "mcp-style static tool manifest", "note": "Read-only data mirroring the Model Context Protocol tools/list and tools/call shapes. This is a static file, not a live MCP server.", "serverInfo": { "name": "laksh-advani", "version": "1.0.0", "url": "https://lakshadvani.github.io/agent.json" }, "tools": [ { "name": "get_profile", "description": "Who Laksh is: role, experience, skills, focus areas and links.", "inputSchema": { "type": "object", "properties": {} }, "result": { "name": "Laksh Advani", "title": "Staff Applied Scientist", "employer": "ServiceNow", "location": "Seattle, WA", "experience_years": "8+", "focus": [ "RL post-training (RLVR, GRPO, KTO, DPO, PPO, process reward models)", "LLM agent reliability and recovery", "Evaluation and LLM-as-a-judge", "AI safety" ], "education": "M.S. Computer Science (ML/AI), University of Colorado Boulder", "experience": [ { "org": "ServiceNow", "role": "Staff Applied Scientist", "dates": "2024–present" }, { "org": "Qualtrics", "role": "Applied Scientist 2", "dates": "2021–2024" }, { "org": "Capital One Labs", "role": "Applied Scientist", "dates": "2019–2021" }, { "org": "Sandia National Labs", "role": "Graduate Research Associate", "dates": "2018–2019" } ], "skills": { "post_training": [ "RLVR", "GRPO", "KTO", "DPO", "PPO", "Process reward models", "RLAIF", "Synthetic data", "LoRA/QLoRA" ], "agents_and_evals": [ "Multi-agent orchestration", "Tool-use RL environments", "Context engineering", "RAG", "LLM-as-a-judge", "Eval harness design", "MCP", "LangGraph" ], "infrastructure": [ "PyTorch", "TorchTitan", "DeepSpeed", "Distillation", "Quantization", "vLLM", "Kubernetes", "Argo", "AWS" ] }, "reviewing": [ "NeurIPS 2026 Workshop IAB", "NeurIPS 2026 Workshop Verify-Agents", "ICML 2026 Workshop FAGEN" ], "links": { "site": "https://lakshadvani.github.io/", "scholar": "https://scholar.google.com/citations?hl=en&user=EBLv9hAAAAAJ", "linkedin": "https://www.linkedin.com/in/lakshadvani/", "github": "https://github.com/lakshadvani" } } }, { "name": "list_publications", "description": "Papers with venue, year and status. Optional topic filter.", "inputSchema": { "type": "object", "properties": { "topic": { "type": "string", "enum": [ "agents", "eval", "reasoning" ] } } }, "result": [ { "title": "From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents", "venue": "ICML 2026 Workshop on Failure Modes in Agentic AI", "year": 2026, "topics": [ "agents", "eval" ], "url": "https://arxiv.org/abs/2606.09863", "key_results": [ "No LLM-judge setup exceeds 0.65 AUROC on tau2-bench", "TF-IDF detectors reach 0.83 (tau2-bench) and 0.95 (AppWorld) AUROC", "3,300x lower latency than judges" ] }, { "title": "Trajectory Guard: A Lightweight, Sequence-Aware Model for Real-Time Anomaly Detection in Agentic AI", "venue": "AAAI 2026 Workshop on Trusted Agents", "year": 2026, "topics": [ "agents" ], "url": "https://arxiv.org/abs/2601.00516", "key_results": [ "F1 0.88-0.94 vs at most 0.69 for unsupervised baselines", "32 ms inference, 17-27x faster than LLM judges" ] }, { "title": "When Small Models Are Right for Wrong Reasons: Process Verification for Trustworthy Agents", "venue": "AAAI 2026 Workshop on Trusted Agents", "year": 2026, "topics": [ "reasoning", "eval" ], "url": "https://arxiv.org/abs/2601.00513", "key_results": [ "50-69% of correct answers contain flawed reasoning", "Distilled verifier: 0.86 F1, 100x faster" ] }, { "title": "Reading the Plan: Multi-Horizon Action Prediction from Frozen LLM Hidden States", "venue": "NeurIPS 2026 Workshops", "year": 2026, "status": "accepted at 5 of 5 workshops", "topics": [ "agents" ], "key_results": [ "75% fewer LLM calls via confidence gating" ] }, { "title": "lakshadvani at SemEval-2026 Task 11: A Neuro-Symbolic Approach to Content-Independent Syllogistic Reasoning", "venue": "SemEval 2026", "year": 2026, "award": "Top 3, Subtask 1", "topics": [ "reasoning" ] }, { "title": "Effective Proxy for Human Labeling: Ensemble Disagreement Scores in Large Language Models for Industrial NLP", "venue": "GEM Workshop @ EMNLP 2023", "year": 2023, "authors": "W. Du, L. Advani, Y. Gambhir, D. Perry, P. Shiralkar, Z. Xing, A. Colak", "topics": [ "eval" ] }, { "title": "C1 at SemEval-2020 Task 9: SentiMix: Sentiment Analysis for Code-Mixed Social Media Text Using Feature Engineering", "venue": "SemEval 2020 @ COLING", "year": 2020, "authors": "L. Advani, C. Lu, S. Maharjan", "topics": [] } ] }, { "name": "list_projects", "description": "Selected production and research systems with headline results.", "inputSchema": { "type": "object", "properties": {} }, "result": [ { "name": "Post-training an agent orchestrator", "summary": "RL environments, weighted verifiers and a turn-level process reward model; KTO warm-start to online GRPO" }, { "name": "Agent recovery loop", "summary": "Localizes failures and resets subagents on long-horizon tasks" }, { "name": "Overnight eval harness", "summary": "Benchmark and live-query evaluation used across the agentic AI organization" }, { "name": "Access-control tripwires", "summary": "Low-latency safety checks on sensitive data access" }, { "name": "Domain LLMs at Qualtrics", "summary": "Fine-tuned Llama 2 and Mistral 7B at 95% of GPT-4 quality and 10x lower cost" }, { "name": "Security ML at Sandia", "summary": "Seq2seq transformers for malware analysis, 10x lower latency" } ] }, { "name": "get_reading", "description": "What Laksh is currently reading.", "inputSchema": { "type": "object", "properties": {} }, "result": [ { "kind": "paper", "title": "Meta²: Precomputing Reusable Knowledge for Online Harness Evolution in Long-Horizon Agents", "authors": "Lingzhi Yuan et al.", "url": "https://www.alphaxiv.org/abs/2610.meta2-online-harness-evolution" }, { "kind": "book", "title": "The Law of the Sublime", "authors": "Robert Greene" } ] }, { "name": "get_places", "description": "Countries visited.", "inputSchema": { "type": "object", "properties": {} }, "result": [ "India", "United States", "Singapore", "Malaysia", "Indonesia", "Canada" ] }, { "name": "get_contact", "description": "How to reach Laksh.", "inputSchema": { "type": "object", "properties": {} }, "result": { "email": "laksh.edu@gmail.com", "preferred_topics": [ "agent reliability", "post-training", "research collaborations" ] } } ] }