Personal prototype / Project 05
AI Search Copilot
Outcome & decision
Built a research-assistant prototype connecting You.com web retrieval with GLM analysis. The deliverable is a search-to-answer integration and a documented workflow for structured comparisons.
- My responsibility
- Search and model API integration; evidence-context and prompt design
- Project scope
- Independent prototype; no production or adoption claim
Evidence basis. Project presentation and supplied implementation account. The code link is retained; current repository access and hosted-demo operation are not established.
Deliverable & scope
Created a prototype that connects a research question, retrieved web information, and a consistent analysis prompt to produce a structured answer. It demonstrates an integration approach for research and comparison tasks.
Problem & constraints
Research comparisons need up-to-date information and a synthesis tied to the material found. The key design constraint was to keep retrieved context distinct from the model’s own prior knowledge.
My contribution & decisions
I connected the search and model APIs, organized retrieved material into an evidence context, and defined the analysis prompt. Separating retrieval, context selection, and synthesis made it easier to inspect where weak answers originated.
Workflow & implementation boundary
The documented workflow is query → web retrieval → material selection → context building → GLM analysis → structured answer.
The presentation also sketches coarse and fine ranking. Without the source code, those boxes should be read as the documented design, not independently verified ranking algorithms.
Limits & next evaluation
No dedicated quality benchmark, production usage, user adoption, or measured time saving is established for this prototype. The two companion search-evaluation studies test separate pipelines and are not performance measurements of this application.