Experience

AI VIET NAM

Software Engineer, Applied AI · Remote, Vietnam · Sep 2025 – Present

Designed retrieval and LLM evaluation experiments comparing semantic chunking strategies on Recall@5, latency, and per-query inference cost, improving retrieval Recall@5 by 20%.
Benchmarked FP8 and INT4 quantization for production LLM inference workloads, profiling latency, throughput, and output-quality tradeoffs, reducing inference cost by 30% while maintaining application quality.
Applied a state-of-the-art ensemble technique to a production AI system, improving reasoning accuracy from 56.16% to 79.18%.
Architected 10 production-grade multi-agent AI systems (RAG, vector databases, FastAPI, Docker, CI/CD, AWS) and mentored 100+ members, with 90% earning AWS Cloud Practitioner certification.
Authored and maintained core AI engineering documentation for 100+ members, standardizing best practices across projects.

Software Engineer, Applied AI · Remote · Apr 2025 – Sep 2025

Trained and evaluated PyTorch models on AWS SageMaker, applying LoRA-based fine-tuning to financial fraud-detection workflows and improving precision by 40%.
Deployed and orchestrated multi-model Python ML services on AWS SageMaker, ECR, and ECS with CI/CD, experiment tracking, model versioning, and monitoring, cutting operational cost by 25% and release cycles from weeks to days.
Defined the technical roadmap and architecture from business requirements, guiding engineers to build a scalable system with reusable components and right-sized resources, cutting development time by 40% and infrastructure cost by 60%.

Software Engineer, Applied AI · Remote · Jan 2024 – Jan 2025

Architected an agentic RAG platform over 500K+ financial documents with similarity search using LangGraph, LangChain, Neo4j, and Pinecone, reducing analyst research time by 60%.
Engineered a multi-agent orchestration pipeline unifying rule-based, retrieval-augmented, and generative components with guardrails and human-in-the-loop review, reducing workflow failures by 70%.
Built a document ingestion pipeline with AWS Glue, Amazon Textract, and PaddleOCR that converts unstructured financial documents into validated LLM-ready datasets, reducing end-to-end processing time by 50%.
Led engineers on agent architecture, running design and code reviews and establishing standards for automated testing, documentation, and design patterns, reducing feature-delivery time by 50%.
Deployed containerized FastAPI-based LLM/RAG services on Kubernetes (Amazon EKS) using Docker and Helm, with horizontal pod autoscaling and rolling updates, achieving 99.9% uptime and zero-downtime releases.

Software Engineer · Remote · Feb 2022 – Dec 2023

Built personalized recommendation models (XGBoost, PyTorch) from user-behavior data, deployed on AWS, and validated variants through A/B testing, lifting CTR by 50% over the previous system.
Built distributed PySpark ETL pipelines to clean, transform, and aggregate large financial datasets, improving data-processing efficiency by 50%.
Automated AWS and GCP infrastructure and CI/CD with Terraform and Tekton, adding automated deployment validation that reduced deployment effort by 60% and sustained 99.9% availability.

Software Engineer · Simi Valley, CA · Jan 2020 – Feb 2022

Developed and deployed a TensorFlow object detection model on embedded aircraft hardware with OpenCV and Scikit-learn, sustaining 60 FPS real-time inference within onboard compute constraints.
Monitored model performance with Grafana dashboards and analyzed detection failures in MATLAB, informing retraining that raised mAP@0.5 from 40% to 50%.
Architected an autonomous UAV flight-control system in cross-platform C# (mobile and desktop) with optimized routing and real-time decision logic, reducing mission execution time by 25%.
Translated operational requirements into scalable system designs with cross-functional teams, streamlining release workflows and reducing application deployment time by 60%.

Research Assistant · Los Angeles, CA · Jan 2018 – May 2019

Processed raw datasets using ETL techniques and trained machine learning model using Scitkit-learn and Tensorflow to improve the object detection and navigation algorithms by 50% in autonomous vehicle.
Deployed optimized NLP, ML, and computer vision models on NVIDIA Jetson AGX Xavier (16 GB RAM) with real-time metrics streaming to AWS CloudWatch, enabling on-device inference at 15 FPS and remote performance tracking.
Integrated camera and sensor inputs with rule-based control logic and ML perception models into a unified decision pipeline, improving navigation accuracy by 60%.