Laksh Advani

About Laksh Advani

Staff Applied Research Scientist focused on translating frontier AI research into scalable production infrastructure. Specializes in the end-to-end lifecycle of Foundation Models, from pre-training and alignment through to high-throughput inference optimization. Expertise lies in architecting Agentic AI systems and Multi-Agent Frameworks (LangGraph) that drive autonomous decision-making at scale. Features a deep, native understanding of AI with over seven years of experience building scalable solutions in both enterprise and research settings.What\'s Been Delivered:• Scaled Agentic systems to handle millions of daily interactions with >90% automated resolution.• Optimized LLM inference speed by 32% through attention optimization without sacrificing accuracy.• Deployed production AI systems handling millions of daily interactions.• Published researcher at EMNLP 2023, COLING 2020, and 2026 on production-ready AI.Core Specializations:• Agentic AI Systems – Multi-agent frameworks (LangGraph), autonomous decision-making, and reasoning.• LLM Optimization – Fine-tuning, RLHF, DPO, KTO, domain adaptation, and inference efficiency (vLLM).• Production ML – End-to-end deployment, MLOps, high-throughput serving, and distributed training.Current Focus: Building next-generation AI systems at ServiceNow.The Sweet Spot: Taking complex AI research and making it work in the real world.

Experience

1. Staff Applied Research Scientist (Agentic Ai)

   ServiceNow

   Jun 2024 — Present · Seattle, WA, US

   Agentic AI & LLMs: Led the design of a production-grade agentic AI system for incident auto-resolution, utilizing ServiceNow’s April 7B model with 4 LoRA heads and advanced RAG strategies. Scaled to millions of tickets with 91% automated resolution and 92% classification accuracy.• Inference Optimization: Reduced Mean Time to Resolve (MTR) by 34% by optimizing vLLM inference pipelines. Achieved <2s end-to-end latency and <40ms hot-swap adapter overhead for high-throughput production environments.• Knowledge Base & Fine-Tuning: Quantified 30% information gaps and reduced deduplication by 23% using Hierarchical HDBSCAN. Integrated these insights to fine-tune a 120B OSS model, significantly enhancing tool-calling and reasoning capabilities.• Multi-Agent Workflows: Designed experimental frameworks using LangGraph and multi-judge alignment systems to validate high-fidelity reasoning trajectories in complex incident scenarios.• Research Publications: Authored 2 papers accepted at 2026 (TrustAgent Workshop) focused on Trustworthy Agents and Trajectory Guards.

Education

• University of Colorado Boulder

  Master of Science, Computer Science

Skills

• Tensor Flow
• Java
• Nltk
• Python
• Cloud Computing
• Scikit-Learn
• Lxc
• Computer Networking
• Big Data Analytics
• Virtualization
• Spark
• Reinforcement Learning
• Decentralization
• Keras
• Machine Learning
• C
• Computer Architecture
• Naive Bayes
• Distributed Systems
• Operating Systems
• Hypervisor
• Support Vector Machine (Svm)
• Data Center Virtualization
• Algorithms
• Docker
• Logistic Regression
• Model Evaluation
• Neural Networks
• Adaboost
• Hadoop
• Openstack
• Pandas
• Data Science
• Lstm
• Topic Modeling
• Data Structures
• Pac
• Blockchain