Deep learning for medical imaging, real-time WebRTC systems, and cloud infrastructure.
Two years at Samsung R&D shipping a bowel-wall segmentation model onto ultrasound machines and moving a live collaboration platform from on-premise installs to AWS — load-tested at 96 concurrent streams.
I build systems where the hard part is invisible — a segmentation model that has to be right inside a clinician's ultrasound, a media pipeline that cannot drop a frame at 96 concurrent streams, infrastructure that scales itself while nobody watches.
Software Engineer with 2+ years at Samsung R&D across deep learning for medical imaging, real-time WebRTC systems and AWS. Shipped a bowel-wall segmentation model onto Samsung ultrasound machines and moved a live collaboration platform from on-premise installs to AWS, load-tested at 96 concurrent streams. BUET Computer Science graduate.
Bangladesh University of Engineering and Technology
B.Sc. in Computer Science and Engineering
Apr 2019 — Jul 2024 · Dhaka, Bangladesh
Years at Samsung R&D
Medical imaging & real-time media
Concurrent streams
Peak WebRTC load test on SonoSync
Awards & honours
Samsung R&D, hackathons, showcases
Public repositories
github.com/saira-yeasmin-333
Building AI and real-time media features that ship inside Samsung medical and consumer products.
Developed, trained and integrated an EfficientDet-D1 detection model and a SegFormer segmentation model that resolve the three layers of the bowel wall and measure each one, giving clinicians far better visualisation. Built and iterated the system inside a WPF simulated environment with a C++ logical backbone and C# visualisation, trained the base model on a curated in-house dataset, and shipped it into the Athena software codebase.
Built a platform that streams an ultrasound machine's output feed in real time to concurrent remote doctors over WebRTC, working across every end of the project. The Spring Boot backend wraps an SFU-architecture Kurento media server that receives and multicasts the stream, then ships containerised on AWS across Route 53, CloudFront, CloudWatch, ECR, ECS, S3 and auto-scaling groups. Also built the agent WPF application and the remote participants' browser front end, and ran load tests at every stage — peaking at 96 concurrent streams.
Extended the AVPlayer media module with RTSP and WebRTC camera streaming, smoothing playback across SmartThings clients.
Full-stack product work on an enterprise Java stack.
Built a full-stack inventory and sales application end to end with Spring Boot, JPA, AngularJS and MySQL.
Medical AI, real-time media, marketplaces and an ERP — production systems and research prototypes, six of them below.
An ultrasound machine's live feed, multicast to remote doctors over WebRTC through an SFU Kurento media server — containerised and auto-scaled on AWS, load-tested at 96 concurrent streams.
EfficientDet-D1 detection plus SegFormer segmentation resolving and measuring the three layers of the bowel wall, trained on a curated dataset and shipped into Samsung's Athena ultrasound codebase.
An RL approach to route planning in road networks that optimises jointly for emissions, noise, travel time and convenience rather than distance alone. Ongoing exploratory research.
A two-sided marketplace matching tutors and students with filters for location, subject and rate — built on PostgreSQL and Express, deployed on AWS.
An ERP for a textile facility covering clothing SKUs, order consignments with timelines and fulfilment status, product catalogues and employee records, on an Oracle core.
An Android app giving commuting women one-tap access to help during emergencies or harassment, backed by Firebase realtime alerts. Champion, Girls' Mobile App Contest.
From CUDA-bound training loops to SDP signalling to auto-scaling groups — the layers I move between day to day.
Ongoing exploratory research
Reinforcement learning for route selection, optimising jointly for emissions, noise, travel time and convenience — treating a route as a policy decision rather than a shortest-path result.