Pulkit Raju Email & Phone Number

Bengaluru, Ka, In

Bengaluru, Karnataka, India

-Engineered an Agentic-RAG (Retriever Augmented Generation) system for Enterprise Search, achieving scalability with over 220 million embeddings (Dense and Sparse Vector fields) using Milvus and Elasticsearch, leading to a 60 percent accuracy increase in end-to-end evaluations– Increased accuracy in enterprise search by implementing advanced prompting strategies like Chain of Thought (CoT) and Few-shot prompting in the answer synthesis (generation aspect) for the RAG solution.– Enhanced already existing data ingestion / indexing pipeline to incorporate Dense and Sparse embedding creation along enabling real-time multi-source synchronization and metadata extraction functionality to improve upon vector search and late re-ranking thus leading to an increase in recall by 40 percent .– Implemented an AI-powered workflow automation engine ( aka Agents ) using function calling architecture, enabling dynamic task orchestration and process automation using different agentic frameworks like DSPy , Langchain and frontier models like Gemini 1.5 Pro, Claude 3 and GPT 4o variants

Bengaluru, Karnataka, India

Bangalore

– Developed a pioneering solution utilizing PaLM-2 LLMs to convert user queries into database queries( Natural language to SQL ), alleviating the hallucination problem inherent in LLMs and boosting accuracy by 45% compared to the previous solution.– Fine-tuned Meta(LLaMA 2 model family) and Google(PaLM ) models by using Prompt Based Fine-tuning and Supervised fine-tuning (SFTP) to adhere to specific instructions and language style requirements for a medical use case, demonstrating a tailored approach to complex problem-solving.– Worked with Nvidia LLMs(NeMo) and HuggingFace ecosystem along with PyTorch using Pruning and Quantization to speed up inference by 30% on CPU while simultaneously reducing overall model size by 50% and in some cases even reducing the size to a third of the orignal utilizing ONNX and TensorRT frameworks for pruning and Quantization of models.

Bhopal

Assisted in organizing debates, competitions, and case studies to promote road safety and awareness amongst youth. Collaborated with a dedicated team to successfully manage events with 40-60 participants each. Utilized social media to conduct awareness campaigns with an average of 200-300 participants. Moderated a debate with 100-200 live viewers to engage the community in promoting the cause.