M. I. Sezer

CurvePoint · Pittsburgh, PA

• Developing real-time gun detection systems using WiFi signal-based sensing and machine learning on edge devices.
• Built and optimized PyTorch inference pipelines for low-latency, real-time detection on edge hardware.
• Developed WiFi signal processing pipelines using CSI and RSSI for sensing and model input generation.
• Implemented system components in Python and C++ across AWS and GCP.
• Worked on end-to-end ML, model integration, edge inference, and production-readiness improvements.

Carnegie Mellon University · Pittsburgh, PA

• Built web-scale knowledge graphs for RAG multi-hop QA.
• Entity extraction, graph traversal heuristics, and evidence aggregation — 15% RAG accuracy improvement.
• Stack: Python · React/Next.js · GCP · MongoDB.

TED University — CADS Research Group · Remote, Turkey

• Member of the Center for Applied Data Science (CADS) research group.
• Agentic AI solutions: planning, tool-use, and multi-agent coordination.
• Prototypes for autonomous task execution, prompt-chaining, and experiment reproducibility.

HubX · Remote, Turkey

• Led team for DaVinci, Tattoo, and Nova — multimodal GenAI products.
• Owned backend ML optimization stack; standardized ML APIs, reducing duplicated infra code by 60%.
• Deployed 13+ production models on AWS/GCP Kubernetes with autoscaling and load-adaptive routing.
• Recommendation engine (Embeddings + FAISS) powering a 1M+ MAU consumer app.

Spacture AI · New York, USA

• Identity-matching platform with ReID and face-matching models for real-time production use.
• GStreamer & NVIDIA DeepStream streaming pipelines for 300+ camera feeds.
• 99%+ face-matching accuracy, 100ms latency via FAISS-based retrieval with graph-based pruning.

METU Research Park Management · Ankara, Turkey

• AI platform analyzing 400+ companies, automating 10K+ monthly reviews.
• Integrated with existing Django + MySQL employee-tracking platform.
• 70% reduction in manual review workload through LLM-based document understanding pipelines.

Orion Robotics · Ankara, Turkey

• Co-founded robotics startup delivering autonomous system prototypes.
• CV models in C++/CUDA optimized for embedded inference — 4× faster, ~50% lower memory usage.
• ROS2 perception and decision pipelines integrating camera, LiDAR, and onboard ML models.