I build production-grade ML and GenAI systems (RAG, fraud detection, MLOps) across AWS and Azure, focused on shipping reliable models that improve real business workflows.
My work spans classical ML, LLM-based systems, data pipelines, and MLOps, with experience delivering solutions in energy, telecommunications, and technology environments—optimizing for latency, quality, and operational efficiency in production.
🎯 Target roles: Data Scientist (Product), ML Engineer, AI Engineer
🎯 Currently focused on:
- Production LLM applications (RAG, agentic systems, evaluation frameworks)
- Scalable ML pipelines and advanced feature engineering
- AI system design for data-intensive products
📍 Location: Raleigh, NC | Open to Remote & Hybrid roles
📌 Status: Actively seeking new opportunities
-
🎓 Master’s in Data Analytics – Northeastern University
-
💼 3.5+ years building production ML systems across consulting and product teams
-
🚀 Deployed 9 production ML pipelines on AWS SageMaker supporting business-critical workflows
-
🏅 Azure Data Scientist Associate
-
🏅 HashiCorp Certified: Terraform Associate
-
📜 IBM Data Science Professional Certificate
-
📜 Google Data Analytics Professional Certificate
These repositories reflect how I design and ship real-world AI systems, not notebook demos.
Production RAG system for customer support over 150K+ product, review, and policy documents.
- LangChain-based retrieval with vector search and source-grounded answers
- Sub-10ms cached responses; cold queries typically under ~1.5s in local benchmarks
- Evaluation harness with 15+ domain-specific test queries
- Dockerized FastAPI backend with Streamlit UI for interactive exploration
Why this matters: Demonstrates how to build reliable, cost-aware RAG systems that can augment or replace human support workflows while remaining debuggable and transparent.
🔗 View Repository: https://github.com/pranshu1921/shopassist-rag
(Includes architecture diagram and reproducible benchmarks.)
Enterprise-style RAG system using agentic reasoning over custom knowledge bases.
- LangGraph orchestration for multi-step reasoning and tool selection
- FAISS vector store for efficient semantic search
- ReAct-style agent with deterministic source citation
- Modular ingestion for URLs and local documents (PDF/TXT/DOCX)
Why this matters: Mirrors how modern AI teams build trustworthy internal copilots and knowledge systems that can be inspected, debugged, and extended.
🔗 View Repository: https://github.com/pranshu1921/Agentic-RAG-Document-Search-System
(Includes architecture overview and evaluation hooks.)
End-to-end ML system emphasizing maintainability and production correctness.
- Data ingestion → feature engineering → training → evaluation → inference
- Clear separation of data, training, and serving layers
- Reproducible experiments with versioning and automated tests
- Dockerized inference service with CI/CD-ready structure
Why this matters: Shows how to take a model from notebook to production in a way teams can operate, extend, and trust.
🔗 View Repository: https://github.com/pranshu1921/ml-production-pipeline
Business-first ML case studies focused on decisions, not just accuracy.
- Churn modeling with cost-benefit thresholding and retention strategies
- Fraud detection under class imbalance with precision-recall tradeoffs
- End-to-end pipelines from raw data to actionable recommendations
Why this matters: Demonstrates how to translate ambiguous business problems into ML systems that drive real decisions.
🔗 View Repository: https://github.com/pranshu1921/applied-ml-case-studies
Applied Machine Learning
Designing predictive models and experiments that solve concrete business problems, from churn and fraud to operational forecasting.
Generative AI & LLM Systems
Building production RAG and agentic workflows with an emphasis on evaluation, reliability, and source-grounded answers.
Production ML Systems
Taking models from notebook to production through deployment, monitoring, CI/CD, and performance tuning.
Data Engineering for ML
Designing SQL-first pipelines, feature stores, and data quality checks that make ML systems reliable at scale.
I gravitate toward work where models become real systems that teams can trust and operate in production.
- Classical ML: Scikit-learn (custom transformers, pipeline optimization), XGBoost, LightGBM
- Deep Learning: PyTorch (NLP, neural models), TensorFlow
- LLM Systems: LangChain, LangGraph, OpenAI API, prompt engineering
- Vector Search: FAISS, ChromaDB, embeddings (OpenAI, HuggingFace)
- Evaluation: Custom metrics, RAGAS framework, A/B testing
- Languages: Python, SQL (complex queries, optimization), R
- Libraries: Pandas, NumPy, Polars
- Feature Engineering: Domain-specific transformations, time series features
- Data Quality: Great Expectations, custom validation frameworks
- Cloud: AWS (SageMaker, Glue, Lambda, S3, Athena), Azure (OpenAI, AI Search, DevOps)
- Orchestration: Apache Airflow, AWS Step Functions
- Infrastructure as Code: Terraform (multi-account setups, state management)
- Containerization: Docker, Docker Compose
- CI/CD: GitHub Actions, automated testing
- Monitoring: Model performance tracking, data drift detection
- Tools: Tableau, Power BI, Streamlit
- Libraries: Matplotlib, Seaborn, Plotly
I believe strong AI engineers:
- Start with problem framing, not models
- Treat data pipelines as first-class systems
- Optimize for maintainability over cleverness
- Measure success using business impact
- Build systems that explain themselves
- Plan explicitly for failure modes
- Advanced RAG architectures (hybrid search, re-ranking, query decomposition)
- LLM evaluation frameworks (RAGAS, domain-specific metrics)
- Fine-tuning and adapting open-source models (Llama 3, Mistral)
- Production monitoring for LLM systems (cost, latency, hallucination detection)
Seeking:
- Data Scientist, ML Engineer, and AI Engineer roles (Remote or Hybrid)
- Contract or consulting projects in AI/ML system design
Happy to discuss:
- RAG system architecture and evaluation
- MLOps best practices and scaling challenges
- Bridging the gap between research and production
Reach me at:
- LinkedIn: https://www.linkedin.com/in/pranshu-kumar
- Email: pranshukumarpremi@gmail.com
⭐ If you find something useful here, feel free to star a repo or reach out.