Harsh Tiwari

I’m an Data Scientist/AI-ML Engineer who architects end-to-end production AI systems for enterprise clients, combining Python, AWS, Docker, and Kubernetes with full lifecycle ownership. I build Python microservices and RESTful APIs with FastAPI and ship conversational AI workflows using GPT-4 Turbo/4o/mini, LangChain, and vector databases like Pinecone and FAISS.

I focus on practical ML engineering: I develop data pipelines, ML inference modules, and async task queues (Celery, Redis) to reduce deployment friction—cutting deployment cycles by 25%. I also build end-to-end AI pipelines for NLP and model deployment with PyTorch, Hugging Face, and FastAPI, improving accuracy by 15% and efficiency by 22% through better ML infrastructure and CI/CD.

In addition to product work, I deliver results through applied research and contract engagements—benchmarking models from BERT and XGBoost to OpenAI and Gemini, and strengthening prediction accuracy by 5–10% with feature engineering and tuning. My medical imaging work used CNNs/Transformers in PyTorch to achieve 92% accuracy and included large-scale processing (10TB+) and graph-based modeling in Neo4j. I bring that same end-to-end, production-minded approach across RAG, semantic search, recommendations, and real-time inference.