Laxman Sharma
AI/ML & Data Systems Engineer
Computer Science graduate (BSc Hons, 2025) from Sunway College Kathmandu with direct experience shipping production-grade NLP translation systems at eSewa (Nepal’s leading digital wallet) and automated ETL pipelines at Dlytica. Passionate about bridging machine learning research with scalable software architecture.
Education & Certification
Work Experience
Worked with the R&D team at Nepal’s premier digital payment platform to build BHASABODH, an intelligent Romanized-to-Devanagari translation model and QA assistant.
- Contributed to training, fine-tuning, and evaluating transformer-based NLP models (BLOOM and BERT variants) on custom language datasets.
- Implemented BLEU score evaluation metrics, data cleaning pipelines, and inference throughput optimization techniques.
- Integrated the fine-tuned translation engine into a high-performance FastAPI web backend paired with a Tailwind CSS frontend.
- Collaborated closely with R&D engineers to test model accuracy and iterate on real-time translation response times.
Completed an intensive data engineering program focused on cloud data warehousing, distributed ETL pipelines, and workflow automation.
- Developed and scheduled Apache Airflow DAGs for automated workflow management and data ingestion pipelines.
- Utilized MinIO object storage for scalable, S3-compatible data lake storage during transformation jobs.
- Gained hands-on experience in cloud platform configuration, database performance tuning, and structured data warehousing.
- Engineered automated Python scripts for cleaning, transforming, and loading high-volume analytics datasets.
Key Projects
Fine-tuned custom BLOOM and BERT transformer variants to translate Romanized Nepali into formal Devanagari. Built a BLEU-evaluated inference optimization pipeline and integrated the model into a FastAPI web interface with interactive QA capability.
Designed, built, and deployed a live Django 4.2 SaaS application helping students discover international universities and scholarships. Hosted on VPS with PostgreSQL, Nginx, Uvicorn, and comprehensive SEO optimization.
Developed an ML risk assessment application using scikit-learn classification algorithms to evaluate clinical patient parameters. Deployed live with an interactive Streamlit frontend for real-time risk predictions.
Architected and executed automated ETL DAG pipelines using Apache Airflow. Integrated MinIO for object storage and Docker for containerized task execution to automate real-time data extraction and processing.