At Vrize India Pvt Ltd, I built backend systems for Tonic and Zaxby’s, turning stakeholder requirements into RESTful APIs and microservices. I traced and resolved 50+ critical bugs, then wrote tests and improved exception handling to help ensure fixes held.
At SRH University of Berlin, I built RAG pipelines over research literature for document classification and information extraction. I tuned retrieval components and checked output against held-out data, with GPT and Claude, then wrapped the workflows in APIs and interfaces.
At Sovrano AI, I reviewed and corrected model-graded agent-task scores before they reached the training pipeline. I also generated ranked preference pairs and wrote annotation guidelines for multi-step workflows.
My projects apply the same habit of validating model output: DataGrid executes generated SQL in a sandbox and logs evaluation runs to MLflow. For my thesis, I analyzed 2,068 logged errors across 560 runs to study failures and repairs in LLM-generated Infrastructure-as-Code.
