Ronit Bakshi

Leadership & Mentorship • Leading recruitment and mentoring two junior engineers, including daily task assignment, code reviews, and technical guidance. Advanced Data Engineering • Developed local ETL setup with Airflow and Spark replacing Databricks for faster testing and iteration. • Built interactive Streamlit + Spark dashboard for querying Delta Lake in AWS S3, and analytics dashboards using Metabase with MongoDB and Trino. Virtual Agent Development • Built virtual agent supporting single and group chats using LangChain, LangGraph, Django, and OpenAI/Groq APIs with ChromaDB RAG pipeline for context-aware retrieval from PDFs and web sources. • Implemented multilingual workflows with seamless language/model switching while maintaining context. • Enhanced user control via DuckDB and SearxNG (Chrome/Brave) for browser-based retrieval, plus conversation templates and prompt orchestration for improved context, relevance, and safety.

Data Engineering • Designed and deployed Databricks ETL pipelines transforming streaming JSON from AWS Kinesis Firehose into Delta Lake with Unity Catalog governance. • Built monitoring dashboards for chat metrics, user engagement, and generation quality. • Prototyped and scaled real-time ingestion pipelines using Kafka and Kafka Connect. LLM Systems • Developed moderation pipelines using BERT and fine-tuned Phi-3 for emotion detection and structured output. • Created prompt templates for enhanced model relevance and safety, plus benchmarking scripts using Locust and multiprocessing. Deployment & Optimization • Configured and deployed inference engines (VLLM, TGI, LM Deploy, MLC LM, Aphrodite) across RunPod, Nebius, and Novita. • Automated autoscaling via Python and RunPod APIs based on queue depth; used Skypilot and Dstack for multi-cloud scaling. • Quantized models to FP8 for optimized memory and throughput; accelerated image generation with Automatic1111 + TensorRT. • Deployed and managed APIs using Docker, FastAPI, PM2, and Nginx for scalable, reliable services.