Ayush Sinha

1. Fine-tuned and quantized 150+ transformer-based LLMs for optimized deployment on Hugging Face, improving inference speed and reducing compute costs. 2. Designed and implemented RAG (Retrieval-Augmented Generation) pipelines using PGVector, FAISS, Pinecone, and Hugging Face embeddings. 3. Built a production-grade Multimodal Generative AI Support Assistant capable of querying PDFs, Excel, CSV, and image-based documents. 4. Integrated LangChain and AutoGen for multi-agent orchestration and real-time streaming responses. 5. Deployed AI inference systems on AWS Lambda, optimizing latency and cloud resource utilization. 6. Integrated Azure OpenAI APIs for automated ticket resolution, improving response efficiency by 60%. 7. Evaluated models using Accuracy, Precision, Recall, F1-Score, and BLEU metrics to ensure production reliability. 8. Reduced manual document search effort by ~35% using semantic search and vector database indexing.

Automated material-certificate data extraction with Automation Anywhere IQ Bot + VBScript, achieving 95% accuracy and saving 4+ hours per batch.

1. Built and evaluated ML/DL models for classification, object detection, and sentiment analysis using Python & TensorFlow. 2. Strengthened practical understanding of data pipelines, NLP, and model evaluation techniques.