Project Context

Throughout this course, you have explored both frontier models (OpenAI, Claude, Gemini) and local open-source models (Ollama, Hugging Face pipelines, and tokenizers). You have also learned how to build interactive user interfaces with Gradio, equip LLMs with custom tools via function calling, and handle streaming outputs.

For your final assignment, you will synthesize these core skills by developing a Hybrid Multi-Model Assistant. This system will dynamically route user prompts based on task complexity: routing basic text tasks (like summarization) to a local open-source model (via Ollama or a Hugging Face pipeline) and routing advanced analytical tasks (requiring external tools) to a commercial frontier model (such as OpenAI's GPT or Anthropic's Claude) with tool-calling capabilities. The entire system must be wrapped in a streaming Gradio chat interface.

Your Tasks

Hybrid Routing Logic (Architectural Setup): Write a routing function in Python that analyzes incoming user prompts. Simple classification rules (e.g., keyword detection or classification via a lightweight local model) should route straightforward text processing or summarization tasks to a local pipeline (Hugging Face or Ollama) while routing complex reasoning or tool-based requests to a frontier API.
Frontier Model & Tool Integration: Implement a custom Python tool (e.g., a mock airline status fetcher, database querier, or system-level tool) and integrate it with your frontier model using OpenAI/Claude function-calling mechanics.
Local Open-Source Pipeline: Load and run a local Hugging Face transformer model pipeline (e.g., for summarization or text generation) or set up a local Ollama integration to handle the simple routed tasks.
Gradio UI & Streaming: Build a Gradio chat interface that displays the conversation. The UI must support streaming responses (yield generator) for both the local and frontier models, and clearly indicate to the user which model (Local vs. Frontier) is currently answering.

Deliverables

Submit a single Python notebook (.ipynb) or script (.py) containing:

The complete, clean Python code with all model initializations, tool definitions, and routing logic.
The Gradio interface code capable of running locally or on Google Colab.
A brief markdown block within the file explaining your architectural routing choice and how the state/context window is managed across models.