Senior AI Engineer

Led the architecture and delivery of production LLM agents and enterprise RAG systems using LangChain, LangGraph, LlamaIndex, FastAPI, Amazon Bedrock, OpenSearch, Qdrant, and tenant-isolated vector indexes. Built orchestration layers for long-running coding and knowledge workflows with typed Pydantic state, explicit tool contracts, checkpointing, retries, and recoverable execution.

Created evaluation suites using golden tasks, LLM judges, deterministic assertions, RAGAS checks, and regression benchmarks. Built multi-tenant SaaS applications and autoscaling inference services with Python, TypeScript, React, Node.js, AWS ECS, PostgreSQL, MongoDB, Docker, Kubernetes, Terraform, SageMaker Pipelines, and Airflow. Instrumented services with Langfuse, trace telemetry, structured logs, and replayable production traces.

Built Python and FastAPI backends for AI-powered customer interview and recommendation platforms. Implemented adaptive text and voice conversations using OpenAI and Anthropic models, Pydantic response schemas, tool calling, conversational memory, human-review workflows, and production safety controls.

Developed secure enterprise assistants with Llama 3.1, Ollama, LangChain, FAISS, and access-aware retrieval. Established MLOps and evaluation workflows with Hugging Face, PyTorch, Docker, MLflow, Azure, LangGraph, LiteLLM, PostgreSQL, and MongoDB. Also delivered Stable Diffusion-based image-to-image toonification models using DreamBooth, LoRA, and SageMaker inference endpoints.

Built automated line-calling systems for professional badminton using synchronized multi-camera video, YOLO object detection, Kalman filtering, epipolar constraints, calibration, and trajectory reconstruction. Developed real-time table-tennis analytics pipelines using pose estimation, action recognition, object detection, and temporal state machines.

Optimized live deep-learning inference with NVIDIA TensorRT, asynchronous video processing, batched preprocessing, and pinned-memory transfer. Created geospatial property-mapping systems with PyTorch U-Net segmentation, satellite imagery, NumPy, SciPy, and post-processing. Managed data preparation, augmentation, experiment tracking, active learning, and model-assisted annotation with Label Studio.

Built NLP baselines for text classification, information extraction, semantic similarity, and document retrieval using tokenization, embeddings, classical machine learning, and neural sequence models. Developed CNN pipelines for image classification, object detection, and segmentation.

Created benchmark datasets, evaluation scripts, experiment harnesses, hyperparameter search workflows, ablation studies, checkpoint management, and reproducible train-validation-test processes. Investigated dataset and model failure modes, including noisy labels, class imbalance, distribution shifts, and sparse examples, and supported model integration through lightweight Python services.

Bachelor's degree in Computer Science.