Kshitiz Regmi

Georgia State University

Atlanta, GA · GPA: 4.2/4.3. Coursework: Distributed AI, Advanced Deep Learning, Digital Image Processing, Advanced Machine Learning, Introduction to Deep Learning.

Georgia State University · Atlanta, GA

• Developing a scalable framework for generating synthetic metagenomic FASTQ replicas for benchmarking analysis pipelines.
• Optimizing large-scale genomic data processing on Arctic Server HPC for reproducibility, scalability, and performance.

Georgia State University · Atlanta, GA

• Taught and supported Python programming, SQL, Pandas, NumPy, SciPy, Matplotlib, Seaborn, data analysis, visualization, and debugging.

Fusemachines Inc. · Kathmandu, Nepal (Client: TIME Magazine)

• Designed the TIME AI platform ML architecture, owned design decisions and product roadmap, and communicated findings to stakeholders.
• Led development and deployment of the TIME AI beta — increased user engagement by 9%.
• Developed a production-grade conversational RAG system using embeddings retrieval, chunking, reranking, and vector DB; generated answers with citations.
• Implemented multi-turn conversation handling with context tracking and follow-up support using Datastore.
• Performed statistical data analysis on 200K articles to define RAG chunking strategy.
• Benchmarked retrieval engines: Vertex AI Feature Store vs OpenSearch, Pinecone, and FAISS — 89% Recall@5 on a 12K+ proprietary benchmark.
• Engineered a shared embeddings and indexing service so RAG, semantic search, recommender, and Document ChatAI reused the same pipeline; eliminated duplicate engineering effort.
• Developed Document ChatAI to help editors ask questions and find information across printed articles.
• Designed, developed, and deployed DeepDive topic extraction and topic-based recommendations for real-time content discovery.
• Developed an unsupervised user segmentation model with clustering on behavioral data; visualized embeddings via TensorFlow Embedding Projector.
• Developed the email marketing engine achieving a 43% unique open rate — won the INMA Global Media Award 2023.
• Developed and deployed a recommendation engine outperforming Google Vertex AI Recommendations — 1.23% CTR lift in production A/B testing.
• Optimized the recommendation engine API to <100ms at 15M+ requests/month with autoscaling and load balancing.
• Deployed ML services on Cloud Run with latency/error monitoring, data validation, and scheduled retraining for drift.
• Led cross-team collaboration with Google Cloud engineers, data engineers, product leads, and stakeholders; contributed to roadmaps, OKRs, and architecture docs.
• Implemented a multi-step multi-agent system on AWS Bedrock for fraud analytics, converting natural language to SQL.
• Developed an LLM-powered agent mapping client data into a common data model via dbt SQL on Trino over Iceberg tables.

Broadway Infosys · Kathmandu, Nepal

• Led Data Science and Machine Learning training for 500+ students — SQL, EDA, data visualization, feature engineering, and model evaluation.
• Mentored students on statistical hypothesis testing, hyperparameter tuning, and error analysis.
• Implemented MLOps pipelines with Git, Docker, and MLflow for experiment tracking and reproducibility.
• Deployed ML models as REST APIs with FastAPI and interactive frontends with Streamlit.

OYA INC · Kathmandu, Nepal

• Engineered an ETL pipeline processing 80K+ user-item records to generate model-ready signals for a collaborative filtering recommendation engine.
• Implemented data preprocessing, data cleaning, and model training and deployment pipelines.
• Deployed real-time inference APIs using Flask, moving the system from prototype to production.