KashmithSamaranayaka
I build production AI systems, from LLM agents to the infrastructure they run on.
Software Engineer with 2+ years building production AI/ML and backend systems: LLM-powered agentic applications, MLOps pipelines, and scalable infrastructure. Open-source core contributor to the Jaseci stack, major developer of the jac-scale Kubernetes plugin, and co-author of two IEEE-published papers on deep learning.
By the numbers
GitHub activity, live
Fetched from the GitHub GraphQL API and revalidated hourly. These are not hand-typed numbers.
1,328
Contributions
past 12 months
2,141
Commits
all-time
476
Pull Requests
all-time
28
Public Repos
on GitHub
2
IEEE Publications
peer-reviewed
2+
Years of Experience
and counting
Selected work
Projects that ship
Agentic AI platforms, generative media pipelines, and published research. Mostly production systems with real users.
Industry · Jaseci Lab
JacHammer
AI agentic software-generation platform
Backend developer on an agentic AI platform that generates and deploys full-stack jaclang applications end-to-end, orchestrating LLM-driven code generation, build, and one-click deployment.
End-to-end app generationIndustry · Jaseci Lab
Tobu Lite
AI image-to-video generation
Built the image-to-video feature for the Tobu platform: a generative media pipeline integrating Google's Veo 3 model that lets users create videos from their photos using selectable templates.
Google Veo 3Industry
Einstein Pro
AI patent generation & document management
Backend developer for an AI platform that generates patents and manages documents, including a RAG-based chatbot over a vector database for fast, accurate retrieval.
Self-Adaptive NILM
Deep learning research → 2 IEEE publications
Designed WaveNet/CNN models with ensemble methods for appliance-level power prediction; applied pseudo-labeling and transfer learning for self-adaptation to appliance aging without retraining, reaching over 97% accuracy on a six-year dataset. Published at IEEE World AI IoT Congress 2024 and in IEEE Access.
Best Presented Paper · IEEE AIIoT 2024Where I've worked
Experience
AI/ML Engineer · Jaseci Lab
Apr 2024 – Present
Ann Arbor, MI, USA · Remote
- Core contributor to the open-source Jaseci stack (550+ GitHub stars) with 30+ merged PRs, as a major developer of the jac-scale Kubernetes deployment plugin and contributor to the jaclang language.
- Developed and maintained backend infrastructure for Tobu, an AI-powered personal memory platform with integrated LLMs serving 1,000+ users; took the system to production fast enough to support the company's $1.5M raise.
- Built scalable PySpark data pipelines for user analysis and feature optimization, and shipped ML features for image analysis and conversation processing.
PythonJacKubernetesLLMsPySparkFastAPIData Scientist · Utech Technologies (Pvt) Ltd
Apr 2024 – Jun 2024
Wattala, Sri Lanka · Remote
- Designed and deployed an end-to-end energy forecasting pipeline with MLflow experiment tracking and model versioning for reproducible, production-ready forecasts.
- Automated ML workflows with Apache Airflow for data ingestion, model retraining, and deployment.
- Built production API services with FastAPI and Docker for real-time energy prediction and dashboard visualization.
MLflowApache AirflowFastAPIDockerPythonApplication Engineer Intern · Synopsys Lanka (Pvt) Ltd
Jan 2023 – Jul 2023
Colombo, Sri Lanka · On-site
- Developed an NLP-based search engine with LSTM models for automated email classification.
- Automated large-file compilation and notification systems using Python, Bash, and CSH scripting.
- Managed 17 processor designs through custom automation scripts for daily performance assessment and resource optimization.
PythonBashLSTMNLP
Education
B.Sc. Engineering (Hons) in Electrical Engineering
University of Moratuwa · Jan 2020 – Jul 2024
Washington Accord-accredited · Dean's List (Semester 2)
Toolbox
Skills & technologies
Languages
AI & GenAI
ML Frameworks
MLOps & Infra
Backend & Data
Research & credentials
Publications
Peer-reviewed work on self-adaptive deep learning for non-intrusive load monitoring.
2024 IEEE World AI IoT Congress · 2024
Self-Adaptive Non-Intrusive Load Monitoring Using Deep LearningM. S. K. Nisakya, S. M. L. Arampola, W. A. Yasodya, S. Kumarawadu, V. Logeeshan, C. Wanigasekara
Introduces a self-adaptive NILM model integrating deep learning with transfer learning and pseudo-labeling, so energy-monitoring models keep working as appliance power patterns drift with age.
Best Presented Paper AwardIEEE Access, vol. 13, pp. 106524–106539 · 2025
Self-Adaptive Deep Learning Framework for Non-Intrusive Load Monitoring: Addressing Aging Appliance Challenges With Transfer Learning and Pseudo LabelingW. A. Yasodya, S. M. L. Arampola, M. S. K. Nisakya, V. Logeeshan, S. Kumarawadu, C. Wanigasekara
Journal extension of the self-adaptive NILM framework, with synthetic data generation and advanced neural architectures achieving high disaggregation accuracy on aging appliances.
Certifications
Machine Learning Specialization
DeepLearning.AI & Stanford University
Machine Learning Engineering for Production (MLOps) Specialization
DeepLearning.AI
TensorFlow: Advanced Techniques Specialization
DeepLearning.AI
Data Structures and Algorithms Specialization
University of California San Diego
Certified Kubernetes Administrator (CKA)
Udemy
AWS Solutions Architect – Professional
Udemy
What's next
Let's build something intelligent.
Open to AI/ML engineering roles, freelance projects, and research collaborations. The fastest way to reach me is email. I usually reply within a day.