I am a senior software engineer who specializes in creating legitimately useful AI applications. I understand LangChain well enough to find it restrictive. I wish the OpenAI chat endpoint would allow partial completions so I could maximize the potential of GPT-4 with Microsoft Guidance. I judge laptops on their ability to run the 70b model.What I Can Do, At a Production Scale Capacity, and Securely:- Build an AI from scratch using PyTorch (please don't make me do this)- Spin up open-source models from HuggingFace and fine-tune them with LORA- Fine-tune models on OpenAI- Use embedding models to populate VectorDBs and have used DeepLake in a production capacity- Integrate these solutions with LangChain, Microsoft Guidance, or a synergistic combination of the strengths of both- Construct a backend using Node.js (Express app) or Python (Flask/FastAPI app) to connect with these AI applications- Design the infrastructure for these complex applications on AWS, with the capability to do so using CDK scripts, ensuring ease of modification and disaster recovery - Develop a frontend for these backends or virtually any other project using React, TypeScript, Material UI, and other modern frontend technologies- Optimize these frontend applications for SEO