Ujjwal Deep

Architected and productionised ML services using PyTorch, Scikit-learn, FastAPI, and MLflow, improving inference performance by ~24% through preprocessing optimisation, evaluation, and model serving improvements. Built RAG and agentic AI workflows using LangChain/LangGraph, embeddings, and vector databases, including ingestion, semantic retrieval, prompt orchestration, and contextual response generation for enterprise use. Optimised LLM inference via quantisation and structured evaluation using LangSmith tracing, reducing inference costs by ~30% while maintaining response quality. Deployed containerised services with Docker/Kubernetes on AWS using CI/CD for reproducible staging/production deployments, automated validation, and serving workflows. Implemented monitoring and evaluation with AWS CloudWatch tracking latency, quality, failures, and model behaviour, reducing investigation cycles by ~20% through structured observability. Delivered maintainable APIs and cloud services in Agile

Developed Python backend services with FastAPI, REST APIs, and PostgreSQL, improving request processing efficiency by ~22% through reusable service components and database optimisation. Integrated ML capabilities using Scikit-learn, Pandas, and NumPy, implementing preprocessing, model validation, performance evaluation, and ML-to-API integration. Automated cloud workflows using AWS Lambda, S3, RDS, IAM, and Terraform, reducing recurring manual operations by ~30% using event-driven processing and repeatable infrastructure provisioning. Containerised backend and ML components with Docker and CI/CD to support consistent builds, automated testing, and reproducible deployment environments. Diagnosed performance issues across application and PostgreSQL with SQL optimisation, logging, debugging, and root-cause analysis, improving troubleshooting efficiency by ~20% across recurring incidents.

Built Python backend modules and REST APIs for e-commerce functionality, contributing to multiple internal features while reducing integration issues by ~20% through structured debugging and code improvements. Implemented PostgreSQL CRUD operations, authentication workflows, and API integrations across customer and product modules to support reliable transaction processing. Improved software quality using Git, unit testing, regression testing, and systematic debugging, shortening defect-validation turnaround by ~30% across recurring release cycles. Collaborated with developers, QA, and business stakeholders in Agile to translate requirements into maintainable backend enhancements, supporting code reviews, testing, and release readiness.