Punit Kunjam

This project demonstrates a comprehensive Procedural Terrain Generator built using C++ and OpenGL. It leverages advanced rendering techniques to create a realistic and immersive environment. The project integrates LLM (Large Language Model) input to allow users to modify terrain parameters dynamically using natural language.

Key Contributions

• Developed and integrated a custom procedural terrain generation system using Perlin noise and advanced shader techniques, incorporating GPT-4 for dynamic terrain modifications based on high-level, natural language prompts, enhancing accessibility for non-technical users.
• Enabled AI-driven terrain customization, showcasing advanced prompt engineering and LLM integration in interactive graphics
• Leveraged GPT-4 function-calling to extract structured data for real-time modification of terrain parameters (e.g., ruggedness, frequency).
• Designed a robust system to parse complex, high-level commands from natural language into precise graphical parameter adjustments, utilizing embeddings and similarity search to interpret varied terrain modification requests.
• Deployed infrastructure to locally run and fine-tune open-source LLMs for low-latency, high-efficiency interaction, optimizing prompt engineering to achieve accurate terrain representations based on vague or abstract prompts. [In Progress]

Tech Stack

• Programming Language: C++, GLSL
• API: OpenGL, OpenAI, Llama
• Libraries: GLFW, GLAD, glm, ImGui, libcurl, nlohmann/json
• Tools: CMake for build management

[This project is part of a product being developed for my startup. Code for the project is available upon request, subject to review based on the nature of the request.]