Shubham Solanki

Built a fully offline, real-time AI assistant that takes live audio input, converts it to text, generates responses using an LLM, and produces human like talking avatar videos. Integrated open-source STT, multilingual LLMs, TTS models, and SadTalker to deliver natural human like avatar responses in 20+ Indian and global languages. Designed the entire pipeline to run efficiently on GPU for fast, low-latency performance.

Developed a Python based tool to automate creation of University TimeTable across 11 schools and multiple departments. Implemented dynamic class scheduling based on room availability, room capacity, and class strength to optimize limited classroom resources. Designed the system to handle complex constraints and inter-departmental dependencies, significantly reducing manual effort and scheduling conflicts.