J
Justin Pettit
About
Detail
California, United States
Contact Justin regarding:
work
Full-time jobs
Starting at
USD180K/year
Finding co-founders
Résumé
Jobs
verified_user
0% verified
- XFounding AI EngineerXUNA AIApr 2024 - Dec 2025 (1 year 9 months)Architected and delivered XUNA Voice/SMS, a production-grade Voice AI platform leveraging LiveKit, Deepgram STT, GPT-4 for reasoning, LangGraph for multi-agent orchestration, Pinecone with Cohere re-ranking for semantic retrieval, ElevenLabs TTS, and Redis for low-latency caching, deployed on GCP Cloud Run to automate end-to-end business phone operations at scale. Fine-tuned and aligned domain-specific LLMs using PEFT + QLoRA on LLaMA 3.1 with Direct Preference Optimization (DPO), achieving higher task adherence, improved response consistency, and cost-efficient inference without full RLHF pipelines. Designed and deployed an agentic AI architecture using LangChain and LangGraph for multi-step orchestration, Model Context Protocol (MCP) for
Principal AI EngineerOpinov8Jan 2022 - Feb 2025 (3 years 2 months)Led the end-to-end development of 20+ agentic AI systems for startup and clients, designing multi-agent orchestration frameworks with LangChain, LangGraph, CrewAI, Haystack and Semantic Kernel to automate workflows across diverse industries like healthcare, e-commerce, enterprise IT, EdTech and finance. Developed multi-modal virtual support agents for B2B SaaS platforms using NLP and LLM techniques, enabling real-time dialogue flow, user behavior adaptation, and task handoff logic across APIs and UI layers using headless browser control and HID emulation. Designed and executed distributed fine-tuning pipelines for open-source LLMs (Llama, Gemma) using QLoRA and Instruction Tuning on SageMaker, accelerating convergence by 45% and reducing do- Senior AI EngineerProgress SoftwareJan 2018 - Dec 2021 (4 years)Developed real-time fraud detection platform for e-commerce transactions leveraging TensorFlow, scikit-learn ensemble models, and AWS SageMaker, achieving 90% reduction in chargebacks and 80% fraud prediction accuracy for Fortune 500 retailers. Built AI-driven customer service chatbot for financial services using Flask, Azure Cognitive Services, and React frontend, supporting multiple languages and reducing average call center volume by 35%. Designed hybrid credit decisioning engine combining regulatory business rules with ML risk scoring via Flask microservices on Kubernetes, leveraging MLflow for model versioning and Weights & Biases for performance monitoring. Implemented product recommendation engine using TensorFlow embeddings with Pos
- ML EngineerAmazonJun 2014 - Dec 2017 (3 years 7 months)Built large-scale recommendation models using DSSTNE and Apache Spark to personalize product suggestions across 100M+ users, contributing to over 35% of e-commerce revenue through improved cross-sell and engagement rates. Developed XGBoost-based ranking models for Search using relevance, semantic, and behavioral signals, increasing top-3 click-through rates by 3.5% and driving measurable gains in customer conversion. Trained multilingual ASR and NLU models using transfer learning to expand voice assistant into five new international markets, achieving production-grade performance across English, German, and Japanese locales. Created real-time fraud detection system using ensemble models on transactional and behavioral features, reducing fal
Education
verified_user
0% verified
- Master of Science in Computer ScienceUniversity of MichiganSep 2012 - May 2014 (1 year 9 months)
- Bachelor of Science in Computer ScienceUniversity of MichiganSep 2008 - May 2012 (3 years 9 months)