Career CoPilot

Career CoPilot is an open-source, retrieval-augmented generation (RAG) chat assistant packaged as a floating widget you can drop into a portfolio or personal site. It answers natural-language questions using your structured career data—not generic web guesses—via Anthropic’s Claude models and a corpus you control.

Source code and releases live on GitHub: github.com/taylorsegell/career-copilot.

Résumés and project pages are usually static. Visitors still ask targeted questions—skills, stack, timeline, impact—and hunting through sections is slow. The goal was a small, embeddable assistant that:

1. Grounds answers in a maintainer-supplied context (RAG), not hallucinated filler
2. Fits visually into an existing site without taking over the layout
3. Keeps lightweight conversation state for a smooth back-and-forth
4. Stays practical to fork: clear integration points, familiar React/Next.js stack

Career CoPilot ships as a floating chat widget plus the plumbing to feed the model curated professional content:

• Your corpus: résumé, skills, projects, and narrative—you decide what the model may cite
• RAG workflow: retrieval + generation so replies align with supplied facts
• Claude via Messages API: configurable prompts for tone and safety boundaries
• Client-side conversation cache: session-friendly history (e.g. localStorage) so returning visitors pick up context

How the pieces fit together:

1. Context packaging: Career data is structured for ingestion—sections map cleanly to what the retriever and prompt expect.

2. RAG pipeline: Queries retrieve relevant chunks from your corpus; the model answers with that material in scope.

3. Widget UX: A corner launcher expands into a full chat surface—responsive, theme-aware, and unobtrusive on mobile.

4. API boundary: Server-side calls to Anthropic keep keys off the client; errors and retries are handled at the integration layer.

5. Persistence: Conversation threads persist locally so multi-turn questions feel coherent without a heavyweight account system.

• Responses draw from the corpus you ship with your fork—not from guessing about strangers’ careers
• Useful for “what stack did you use on X?” or “summarize your ML governance work” when that lives in your files

• Natural-language questions and follow-ups
• Tone tuned for professional, approachable copy

• Floating entry point; minimal invasion of page chrome
• Responsive layout and motion that respect reduced-motion preferences where configured

• Next.js + React baseline
• Straightforward path to swap corpus, styling, or model settings for your own brand

Career CoPilot turns long-form career material into something visitors can query instead of only scroll. Maintainers get:

• Faster discovery of skills, roles, and projects through dialogue
• A credible OSS demo of RAG + Claude wiring for hiring managers and collaborators
• A reusable widget others can adapt for their portfolios

Repository: clone or fork Career CoPilot on GitHub, follow the project README for setup, and wire your own corpus.

Live example: this portfolio still ships Ask Taylor in the bottom-right corner—the same RAG-style widget pattern, customized here while Career CoPilot is the open-source project others can fork.

Example prompts (when the corpus includes similar detail):

• “What are your strongest AI or data architecture themes?”
• “Walk me through a notable project from your portfolio.”
• “Where have you applied RAG or governance patterns?”

Career CoPilot is meant to stay maintainer-owned: your data, your deployment, your prompts—with the heavy lifting of widget + RAG scaffolding shared in the open.