Software Engineer

Taharina
Tasnim

Building AI that ships to the real world

Computer vision, on-device deep learning, and ML deployment for enterprise hardware at Zebra Technologies. M.S. Computer Engineering, Stony Brook University.

View Projects Get In Touch
Currently
Software Engineer II
Zebra Technologies · Holtsville, NY
Focus
Computer Vision · On-Device ML · Android
Education
M.S. Computer Engineering, Stony Brook
Research
DNABERT-2 · Human Genome · Clustering
01

About Me

I'm a Software Engineer at Zebra Technologies, where I design, train, and deploy production computer vision and deep learning models onto enterprise Android hardware used at scale across global warehouses and logistics operations.

My work spans the full ML lifecycle — from architecting model pipelines and fine-tuning pretrained networks, to Zebra AI Suite SDK integration, performance benchmarking, and on-device TFLite deployment. I thrive at the intersection of rigorous research and real-world engineering constraints.

Furthermore, I have experience in applying large language models (LLMs) in industrial applications.

98%
CV Accuracy
25%
Scan Lift
1M+
Users Reached
AI & Machine Learning
Frameworks
PyTorch · TensorFlow · Keras · HuggingFace · YOLOv8 · OpenCV · scikit-learn
Deployment
TFLite · Android SDK · Zebra AI Suite · INT8 / FP16 Quantization · On-Device Inference
Languages
Python · Java · Kotlin · C · C++ · C# · TypeScript
Infrastructure
AWS · GCP · Docker · Kubernetes · Jenkins · Git · Perforce · Gerrit
02

Experience

Jan 2025 — Present
Zebra Technologies
Holtsville, NY
Software Engineer II
  • Designed and trained a PyTorch CV model achieving 98% accuracy detecting expiration dates on packaging; deployed to production for enterprise compliance automation.
  • Led end-to-end Zebra AI Suite integration — model integration, SDK integration, and performance benchmarking — into Android DataWedge applications on Zebra scanners.
  • Collaborated cross-functionally with BU managers to resolve critical library blockers, driving scanning accuracy to 95%+.
May 2024 — Aug 2024
Zebra Technologies
Holtsville, NY
Intern — Mobile Computing Engineer
  • Built an image qualification algorithm for barcode decoding, optimized for real-time Android performance; improved scan success rate by 25%.
  • Designed a fruit-quality assessment solution using signal processing and CV; awarded Top 3 at company-wide innovation event.
May 2019 — Aug 2021
Samsung R&D Institute
Dhaka, Bangladesh
Software Engineer I
  • Delivered WearOS and Tizen Galaxy Watch features (Settings, PPT Controller, Compass, Camera Controller) in Java and C++.
  • PPT Controller shipped on-time for Watch 5 series, reaching 1M+ users globally.
  • Resolved 120+ software defects across Agile sprints with post-release maintenance accountability.
03

Selected Projects

Project 01 — Featured
ParcelScan — Damaged Packaging Detection
EfficientNetV2B0 · YOLOv8 · TFLite · Android · TensorFlow
End-to-end warehouse demo application detecting damaged packaging in real time. Two-model pipeline: a fine-tuned EfficientNetV2B0 classifier and YOLOv8n object detector, both quantized and exported to TFLite for on-device Android inference. Custom 500-image dataset augmented to 350 per class, with INT8 quantization producing a ~20 MB model. Custom NMS logic for the YOLOv8 [1, 84, 8400] output tensor implemented in Android.
Project 02
Bayesian Classification on Cardiotocogram Data
Python · Bayesian Learning · scikit-learn
Applied Sequential Bayesian Learning with Gaussian basis functions. Achieved 51.6% faster training versus a ridge regression baseline on the Cardiotocogram medical dataset.
Project 03
Denoising Diffusion Probabilistic Model
PyTorch · DDPM · Generative Modelling
Comprehensive survey and prototype implementation of Denoising Diffusion Probabilistic Models for generative image modelling from statistical first principles.
Project 04
Local Feature Descriptor & Model Fitting
Python · OpenCV · Computer Vision
End-to-end CV pipeline combining Gaussian filtering, Harris Corner Detection, and normalized gradient direction histograms for robust edge and corner detection.
04

Independent Research

Ongoing
2025 — Present
Independent · Remote
DNABERT-2 Genomic Analysis & Clustering
Transformers · Bioinformatics · HuggingFace · Python
Investigating the capability of DNABERT-2 — a pre-trained transformer model for genomic sequences — to generate contextual embeddings from human genome data. Applying unsupervised clustering (K-Means, UMAP) to identify functional and structural patterns across genomic regions. This work extends a long-standing interest in bioinformatics, rooted in an undergraduate thesis on alignment-free phylogeny reconstruction, into the era of large language models applied to biological sequences.
05

Publications

01
Automated Cost Reduction While Using Multiple SIMs in a Single Mobile Device
P. Chakraborty, S. Arabi, A. Siddiqua, T. Tasnim, et al.
4th NSysS · 2017 · IEEE
View ↗
02
Service-Hub: A Better Approach for Developing the System of Online Marketing for Daily Services
A. Bhattacharjee, T. Tasnim, et al.
4th NSysS · 2017 · IEEE
View ↗
03
An Alignment-free Method for Phylogeny Reconstruction Using Maximum Likelihood
T. Tasnim · Undergraduate Thesis, BUET
BioRxiv Preprint · 2019
View ↗
06

Get In Touch

Open to Software Engineer, AI Scientist, ML Engineer, and Computer Vision roles. Always happy to discuss research, on-device ML, genomics, or collaboration opportunities.

Send an Email
✉
Email
tasnim.taharina@gmail.com
→
</>
GitHub
github.com/minionpika
→
in
LinkedIn
taharina-tasnim
→
◎
Google Scholar
View Publications
→