Upgradable AI Engineering Tutor

I’m looking to have an AI-powered chatbot built from the ground up that can freely search, learn, and reason through complex engineering homework and typical college-level problem sets while also acting as a general research companion. Engineering is the first domain I need it to master, yet the codebase should be cleanly modular so I can plug in new subject models and extra data sources later on.

Core capabilities I expect from the first release
• Conduct literature-style research and summarise findings on demand
• Walk me through multi-step engineering calculations and proofs
• Tackle full homework questions end-to-end, showing the steps so I can learn from them
• Handle routine “college class” queries—definitions, derivations, examples—without prompt limits or hidden filters

System requirements and deliverables
1. A working chatbot (preferably Python) that runs locally through a web or desktop interface and calls a large-language model you help me select (e.g., open-source LLaMA, GPT-J, or other license-compliant alternatives).
2. A plug-in or micro-service architecture so future models, vector databases, or external tools (like Wolfram|Alpha, MATLAB, or Jupyter kernels) can be added with minimal refactor.
3. Clear, step-by-step deployment documentation covering hardware specs (CPU/GPU, RAM, storage), containerisation options (Docker, Kubernetes if warranted), and any cloud/on-prem trade-offs so I can decide whether to host on my own workstation or a small server cluster.
4. A short roadmap that outlines how I could fine-tune the model on my own datasets or swap in more powerful checkpoints as they become available.

Timing is flexible—quality and extensibility matter more than speed—so feel free to propose an approach that balances performance with long-term maintainability.