Vedant Rajpurohit
AI models need backend. I build both.
I build AI systems by understanding what happens inside the model, not just by calling APIs.
My work focuses on LLMs, fine-tuning, RAG pipelines, agentic workflows, voice AI, and production ML systems. I like exploring how models learn, how tokens become outputs, how reasoning behavior improves, and how fine-tuning methods like SFT, DPO, and LoRA can shape a model for real-world use.
I enjoy working close to the core of AI systems - training, evaluating, improving, and deploying models that are not only powerful in demos, but reliable in production.
Open to AI/ML collaborations, research projects, and meaningful technical conversations.
Advanced LLM systems, agent workflows, and production ML evaluation.
A computer-vision virtual painter that turns hand gestures into brush strokes, color selection, and canvas controls.
A Llama 3.1 8B model fine-tuned with Unsloth and TRL to translate natural-language questions into SQL queries.
A hands-free voice pipeline that listens through VAD, transcribes with Whisper, generates an LLM reply, and renders a lip-synced avatar video.
A multilingual real estate assistant for FAQ chat, document RAG, web-link retrieval, image chat, and English/French/German responses.
A computer-vision virtual painter that turns hand gestures into brush strokes, color selection, and canvas controls.
A Llama 3.1 8B model fine-tuned with Unsloth and TRL to translate natural-language questions into SQL queries.
A hands-free voice pipeline that listens through VAD, transcribes with Whisper, generates an LLM reply, and renders a lip-synced avatar video.
A multilingual real estate assistant for FAQ chat, document RAG, web-link retrieval, image chat, and English/French/German responses.