M. I. Sezer
i@imsezer.comAI Engineer - Pittsburgh, PA
ML Systems Engineer · Edge & Multimodal · CurvePoint
I build production ML systems on top of edge, cloud and robotics
My path into computing started with a 2015 MIT video of a robot learning to drift.
Trajectory
Where the systems shipped
CurvePoint · current · Pittsburgh
Real-time gun detection over WiFi (CSI/RSSI) — low-latency PyTorch edge inference, C++/Python on AWS & GCP.
Carnegie Mellon · Research Collaborator, ML
Web-scale knowledge graphs for multi-hop RAG — entity extraction, graph traversal, evidence aggregation.
HubX · Software Team Lead, GenAI
Led DaVinci / Tattoo / Nova multimodal products · 13+ production models on K8s autoscaling · −60% duplicated infra.
Spacture AI · MLE · New York
Real-time ReID + face-matching across 300+ camera feeds — FAISS retrieval with graph-based pruning.
METU Research Park · AI Systems
LLM document-understanding platform analyzing 400+ companies, automating 10K+ monthly reviews.
Orion Robotics · Co-Founder, ML/SW
Embedded computer vision in C++/CUDA + ROS2 perception (camera + LiDAR) — ~50% lower memory.
TEKNOFEST · AI in Transportation
1st place — autonomous-vehicle computer vision, with minimal PyTorch experience at the time.
Selected work
All projectsgraphmem — Knowledge-Graph Memory
Shared persistent memory for AI agents. An MCP server and CLI combine semantic recall, keyword search, and typed graph traversal in Oracle Autonomous Database.
- Python
- Oracle ADB
- MCP
- ONNX
- Knowledge Graphs
SPIRE — Structure-Preserving Retrieval
Research on retrieving evidence from HTML while preserving its document structure. Sentence paths, contextual filtering, and token budgets keep evidence precise and citable.
- HTML
- BGE
- RAG
- Contextual Filtering
Eflatun AI — Aerial Traffic Detection
Bird’s-eye-view vehicle and pedestrian detection, a published training dataset, and a first-place TEKNOFEST team result. A companion public demo makes predictions inspectable.
- Python
- PyTorch
- YOLO
- Computer Vision
- OpenCV