Akhila Boddu

Implemented enterprise-grade LLM safety controls by combining AI, a custom policy rule engine, and vector-based filtering, ensuring compliant, secure, and toxicity-reduced model outputs for production environments. Built embedding pipelines using OpenAI Embeddings, domain-tuned representations from HuggingFace, and Sentence Transformations for accurate semantic search and document retrieval. Developed robust RAG ecosystems by orchestrating retrieval and generation with LangChain, applying vector indexing through Pinecone for real-time search and FAISS for scalable offline/on-premise retrieval. Executed model finetuning with LoRA, QLoRA, and RLHF across GPT-4, LLaMA, and T5 to improve accuracy while reducing inference cost. Enhanced retrieval quality with ranking models like ColBERT and DPR, establishing retrieval consistency checks to strengthen grounding and reduce hallucinations. Strengthened NLP preprocessing and knowledge extraction with SpaCy and implemented entity extraction, tex

Implemented enterprise-grade LLM safety controls by combining AI, a custom policy rule engine, and vector-based filtering to ensure compliant, secure, and toxicity-reduced outputs in production. Built embedding pipelines using OpenAI Embeddings, domain-tuned HuggingFace models, and Sentence Transformers for efficient semantic search and document retrieval. Developed robust RAG ecosystems by orchestrating retrieval and generation workflows with LangChain and Pinecone for real-time search, and FAISS for scalable offline retrieval. Performed model fine-tuning with LoRA, QLoRA, and RLHF across GPT-4, LLaMA, and T5 to improve accuracy while reducing inference cost. Enhanced retrieval quality with ranking models like ColBERT and DPR, and established retrieval consistency checks to strengthen grounding and reduce hallucinations. Strengthened NLP preprocessing with SpaCy for better upstream context and performed knowledge extraction. Implemented multi-agent AI systems using LangChain Agents an

Fine-tuned LLMs using Hugging Face Transformers, PyTorch, and PEFT/LoRA; applied to new tasks, reduced training costs, and delivered more accurate summarization and reasoning. Built retrieval-augmented pipelines by orchestrating document flows with LangChain and indexing embeddings through FAISS/Pinecone to ensure grounded, contextually accurate responses. Trained and evaluated models with scikit-learn for baselines, then scaled to deep learning solutions. Created advanced NLP and extraction workflows using TensorFlow and OCR/ONNX models to pull structured information from unstructured text and images. Managed the full ML lifecycle with MLflow to track experiments and SageMaker Pipelines to automate training, validation, and deployment. Automated training and data workflows with Airflow; cleaned and engineered large datasets with Pandas; processed big data with Spark on EMR; deployed self-contained services with Docker and managed scalable inference on Kubernetes.