4+ years of industry experience, including internships, across 3D computer vision, robotics perception, GPU work and AI infrastructure. Also see full-time roles and internships & research.
Roles
-
Computer Vision AI/ML Engineer
- Official Python SDK for Actian VectorAI DB. Lead developer and release owner of actian-vectorai-client, which shipped at v1.0.0 on PyPI on the product's launch day. It provides sync and async clients over gRPC and REST, TLS, retries, API-key authentication, a filter DSL and hybrid fusion.
- Engineering quality. SDK code is generated from shared Protocol Buffers definitions, and parity tests keep the client in step with the server.
- Benchmarking. Benchmarked VectorAI DB against leading open-source vector databases with VectorDBBench. At launch, Actian reported more than 22× their throughput at 10M vectors.
- MCP agents. Designed and built an MCP server for B2B prospecting with OAuth 2.1, per-tenant isolation and an SSRF guard. After a failed customer demo, I rewrote the tool descriptions; GPT-4o now picks the right tool for 31 of 31 real sales phrases, and 491 tests pass.
- Legacy ERP to GraphRAG. Connected a decades-old construction ERP to the agent platform in three weeks. Inferred 5,108 relationships across 43,363 columns from undeclared keys, and cut warm schema calls from about 2 s to 18 ms.
- Inference and NVIDIA. Vision and LLM inference on NVIDIA Triton, vLLM, ONNX and TensorRT. Integrated the NVIDIA VSS blueprint and NeMo retrieval.
-
Computer Vision AI/ML Engineer Intern
- Analyzed construction progress from 360° site video on NVIDIA VSS and NeMo, using vision-language models, DINOv2/CLIP keyframe extraction and VideoRAG. Built digital twins in NVIDIA Omniverse.
- Built a Gemini 2.5 Flash analyzer that checks site video against Primavera P6 schedules and produces reports, structured JSON and annotated video.
- Prototyped robotic site perception with ORB-SLAM3 localization and TensorRT models on NVIDIA Jetson Orin.
-
3D AI Lead Engineer
- Led the core architecture for text-to-3D, image-to-3D and video-to-3D generation (NeRF, photogrammetry), producing XR-ready assets.
- Built multi-GPU training and inference (PyTorch DDP) with Docker GPU hosting on a decentralized network, which cut compute cost by 30% in real-world testing.
- Added data masking so personal GPUs could process client workloads privately in exchange for tokens.
-
Computer Vision Engineer
CV Intern, then Software Engineer, then Computer Vision Engineer.
- Built a 360° video-to-3D pipeline for AEC clients combining Mask R-CNN, OpenMVG/OpenMVS and a 3D geometric voting algorithm, with 90% background-noise reduction.
- Made UV-texturing projection 15× faster with a multithreaded geometric formulation, so 10M+ point scans went from hours to minutes.
- Fused iPhone LiDAR, Intel RealSense and IMU data for SLAM: 25% better trajectory accuracy and 40% faster point-cloud processing.
- Ran real-time 3D capture on NVIDIA Jetson TX2 and iOS (ARKit) at 10–15 FPS.
-
Research Project Intern
- Developed UIE-Net, a GAN that enhances underwater images for autonomous underwater vehicles, with a 60–75% visibility improvement. Accepted at GC-RDCT 2021.
-
Project Trainee
- Built a computer-vision tool that scores micro-suturing performed by trainee neurosurgeons.