Shivam Gehlot

I am an AI/ML-focused software engineer with hands-on experience building NLP systems and production-ready ML applications. I specialize in retrieval-augmented generation (RAG), transformer fine-tuning, and end-to-end pipelines that move models from research to deployed services.

I designed and deployed NyayAI, a legal RAG chatbot for Indian criminal law using LangChain, Qdrant, Llama 3.3, and Streamlit, and I implemented semantic caching and reranking to reduce latency and redundant LLM calls. I built a tweets sentiment analysis web app, benchmarking TF-IDF + Logistic Regression, LSTM, and a fine-tuned BERT model that achieved 86% validation accuracy and was deployed via Streamlit.

My video summarization pipeline converts YouTube or uploaded videos into structured summaries using Whisper ASR, Qwen GGUF, and adaptive summarization strategies (Direct, Hierarchical, Map-Reduce) to manage context and token usage. I focus on practical engineering choices—chunking, caching, and efficient model orchestration—to deliver reliable user experiences.

I pair technical depth in Python, ML frameworks, and deployment tools with a commitment to mentoring and community: I ran a 4-week robotics workshop and organized IEEE events. I seek roles where I can drive real-world NLP products and improve model performance and latency in production environments.