AI Engineer

Role

• Built a proof-of-concept continual instruction tuning pipeline for Llama-3.1-8B using SFT and PEFT with DeepSpeed, and successfully reproduced the results of the TRACE continual learning benchmark (2023) as a first step toward testing on real-world data.
• Built RAG Agents using LangChain and Ollama with word and topic guardrails, achieving a 15% improvement in response accuracy through tunable contextual grounding and relevance scoring.
• Developed LLM instruction tuning pipelines with Dagster, AWS S3, SQS, DynamoDB and TorchServe.