Tanwi Kumari

About Tanwi Kumari

I am a Software Developer with 3 years of professional experience, and for the last 2 years I have been deeply focused on Generative AI and Large Language Model (LLM)–based applications.During this journey, I explored multiple foundation models (GPT, Claude, Mistral, Sonar, Qwen, kimi, deepseek, LangChain-based pipelines) and AI platforms such as Amazon Bedrock, Hugging Face, GCP Vertex AI, and more.Developing features that improve and personalize the user’s search experience using Generative AI.My work primarily involves building end-to-end GenAI solutions, including: • RAG (Retrieval-Augmented Generation) systems with document loading, HTML content extraction, chunking, embeddings, and vector search. • Designing pipelines using FAISS, semantic search, and custom context mixing for high-accuracy answers. • Building AI-powered applications that combine model knowledge + custom dataset knowledge.I have also built multiple AI-driven products:1. AI Chatbot for Enhanced Search ExperienceA full conversational AI system where users can: • Ask any query and continue the discussion naturally (multi-turn) • Get answers enriched with images, source citations, charts, tables, and structured results • Explore follow-up queries auto-generated by the model • Receive hybrid answers using both vector-search context + model knowledgeThis is designed to make search more interactive, visual, and intuitive.2. AI Summarization System + Chrome Extension • Built a summarizer for webpages and selected text • Integrated with a Chrome Extension to summarize any open tab • Added chatbot-style conversational UI with: • Continue conversation • Stop generation • Start new query • Streaming responses3. Structured Output Generation Using LLMs • Converts any query into a strict, schema-based JSON using LLMs • Works with custom domain data + validation logic4. Additional AI Tools • Automated product review generation • Context-aware chatbot flows • Search experience enhancements using generative AI • AI-powered content extraction toolsI’m continuously experimenting with advanced LLM streaming, embeddings, RAG enhancements, prompt design, and production-ready AI architectures to build smarter, faster, and more reliable GenAI systems and make search smarter and more meaningful.Always exploring more ways to make AI useful in real-world applications.

Experience

1. Senior Software Developer

   Hyperdart

   Aug 2025 — Present

   Developing AI generated schema-based structured output generators for the user query understanding using our own provided data to enhance the user search experience.Created an MCP server where registered multiple resources and tools to be used by user to do Resource/domain classification based on user query intent and then generate query object structured output using provided schema for user asked query.Exploring different llm models with its features to reduce the hallucinations and enhances the timing and accuracy of response.Learning more about prompt engineering according to our use case to return the expected accurate output to the user without any hallucinations.

Education

• S.L DAV Public School

  10th

• Jaypee Institute Of Information Technology

  Bachelor of Technology - BTech, ECE

  2018 — 2022