TEJA. V

I’ve been working at the intersection of ML, GenAI, and platform engineering—building systems that go from experimentation to stable production. I design end-to-end ML and LLM workflows covering data ingestion, feature pipelines, training, deployment, and monitoring, with a strong focus on reproducibility and faster iteration. I’ve modernized legacy analytics by migrating R-based pipelines to Python on Databricks, reducing execution time by ~60% and improving maintainability. I’ve also built scalable PySpark pipelines to process IoT and application data, improving data freshness and reducing latency for downstream models. On the serving side, I’ve developed production APIs using Flask and FastAPI, and deployed containerized services using Docker and Kubernetes. I’ve set up CI/CD pipelines with GitHub Actions, Azure DevOps, and Jenkins to automate builds and reduce release cycles from weeks to hours. For GenAI, I’ve built RAG-based systems using Azure OpenAI and Azure AI Search, handling chunking, embeddings, vector search, and re-ranking to improve grounding. I’ve also implemented conversational workflows with prompt templates, guardrails, citations, and session memory. To ensure quality, I’ve built evaluation pipelines using golden datasets and tracked latency and cost in production. From an MLOps side, I’ve used MLflow for experiment tracking and versioning, and set up observability for drift, performance, and auditability. I’ve worked across Azure, AWS, and GCP, using services like Lambda, Step Functions, SageMaker, and Vertex AI for real-time scoring and orchestration. Day to day, I collaborate with data engineering and product teams to translate business needs into deployable ML solutions, while keeping systems well-documented and easy to extend.

I’ve been working mainly on backend systems in Python, building services that are actually used in production and easy to maintain over time. Most of my work involves designing clean, modular code and exposing functionality through REST APIs using frameworks like Flask and Django. On the data side, I’ve handled a lot of tabular processing using Pandas and NumPy—everything from transformations to building internal analytics tools for tracking usage and web traffic patterns. I’ve also worked closely with databases like MySQL, PostgreSQL, and MongoDB, designing schemas, optimizing queries, and using ORM layers to keep the codebase clean and maintainable. I’ve built and deployed Python microservices on AWS (EC2, S3, RDS), focusing on scalability and low-latency APIs. To keep environments consistent, I’ve containerized services using Docker and set up CI/CD pipelines with Git, Jenkins, and GitHub to automate builds, testing, and deployments. A good chunk of my work has also been around automation—writing Python and Bash scripts to handle repetitive backend tasks, data conversions, and integrations. I’ve worked with messaging systems like RabbitMQ for async workflows and even automated cluster setup/config when needed. Day to day, I’m involved across the full SDLC—from requirements and design discussions to deployment and production support. I work closely with product, ops, and infra teams to make sure what we build actually solves the problem and runs reliably once it’s live.

I’ve been working on production ML systems and structured data problems, handling everything from raw data to deployed models. I use Python (NumPy, Pandas) for data processing, feature engineering, and exploratory analysis, and build classification, regression, and time-series models for use cases like demand forecasting, inventory optimization, and customer behavior analysis. I’ve built end-to-end ML pipelines covering data ingestion, training, evaluation, and inference, making sure models are reproducible and reliable across environments. For modeling, I’ve used both deep learning (TensorFlow, PyTorch) and classical approaches like XGBoost, CatBoost, and ensemble methods, depending on the problem. On the GenAI side, I’ve built RAG pipelines using FAISS, BM25, DPR, and KNN to improve grounding and reduce hallucinations. I’ve also worked on LLM applications with prompt design, session memory, and safety filtering, and optimized models using LoRA/QLoRA and MoE approaches. I’ve evaluated models across quality, latency, and cost to pick the right fit for production. I’ve set up evaluation workflows using golden datasets and A/B testing, and used MLflow for experiment tracking and versioning. On the data side, I’ve built PySpark pipelines and migrated legacy R workflows to Python on Databricks. I’ve also worked on scaling training with DeepSpeed and FSDP, using mixed precision to reduce compute cost. Day to day, I collaborate with product and engineering teams to turn business problems into deployable ML solutions, while keeping systems well-documented and maintainable.

I’ve been working across backend engineering and machine learning, mostly building systems that actually make it to production and stay stable there. On the backend side, I built Python-based APIs with caching using Redis, which helped bring down response times and handle higher traffic without scaling costs too aggressively. I’ve also worked quite a bit with data pipelines using Pandas and MySQL, pulling and shaping datasets so they’re actually usable for modeling. On the ML side, I’ve owned end-to-end pipelines—from ingestion and feature engineering all the way to training, evaluation, and inference. A big focus has been making these pipelines repeatable, so experiments don’t break when you move from dev to prod. I’ve used models like XGBoost, LambdaMART, and factorization machines for ranking, recommendations, and forecasting use cases. For example, I worked on pricing optimization and demand forecasting models that improved margins and reduced waste, and also built supply chain models that helped maintain SLAs while cutting operational costs. I’ve also been getting deeper into GenAI—building RAG-based systems and LLM workflows where grounding and evaluation actually matter. That included setting up prompt versioning, retrieval tuning (KNN/embedding-based), and A/B testing pipelines to measure real impact in production instead of guessing. From an MLOps perspective, I’ve put effort into observability and governance—tracking drift, latency, and cost, and making sure everything from datasets to model outputs is traceable. I’ve used MLflow for experiment tracking and reproducibility, and containerized services with Docker so deployments are consistent. Most of these services were deployed on AWS/Azure with proper CI/CD pipelines, secure APIs, and controlled rollouts. Day-to-day, it’s a mix of coding, debugging production issues, and working with product and data teams to translate business problems into something we can actually ship.