GoCareerPath

AI research agent that finds 4 well-paying, automation-complementary roles you can pivot to without resetting your career, and gives step-by-step resources to get there.

---

Overview

GoCareerPath is a research-first AI agent that scans web search and knowledge sources, compresses findings, and outputs a focused final report listing four practical career pivots tailored to the user's background. It prioritizes roles that are complementary to automation and provides step-by-step learning and job-transition resources.

Core capabilities:

• Multi-provider LLM support via init_chat_model() (configurable in LangGraph Studio).
• Structured research pipeline: summarization → research → compression → final report.
• Pluggable search API support (default: Tavily) and full MCP compatibility.
• Deployable as a web app (Vercel) with LangGraph Studio UI for control.

---

Tech stack

• Node.js, Next.js, React
• LangChain-style orchestration (LangGraph)
• LLM providers: OpenAI (gpt-4.1, gpt-4.1-mini), Anthropic/Gemini, other providers via init_chat_model()
• Search: Tavily (default), native web search integrations for Anthropic/OpenAI, MCP servers
• Hosting: Vercel (web app)

---

How it works (high level)

1. User input — background, skills, role preferences.
2. Search / Research — research agent queries configured search tools (Tavily, web search, MCP).
3. Summarization — summarizer model digests search results (default: openai:gpt-4.1-mini).
4. Research pass — primary compound research using designated research model (default: openai:gpt-4.1).
5. Compression — compress findings for concise evidence (default: openai:gpt-4.1).
6. Final report — structured output: 4 role suggestions + rationale + step-by-step resources (default: openai:gpt-4.1).

Note: models must support structured outputs and tool-calling.

---

Default model roles

• Summarization: openai:gpt-4.1-mini — condense search results
• Research: openai:gpt-4.1 — power the search agent and deeper reasoning
• Compression: openai:gpt-4.1 — compress intermediate findings
• Final Report: openai:gpt-4.1 — generate the actionable final report

You can override these with any provider supported by init_chat_model() in LangGraph Studio.

---

Search API & Tooling

• Default: Tavily search API
• Native web search adapters for Anthropic and OpenAI
• Full MCP compatibility for custom servers and enterprise data sources
• Add new search tools by implementing the search adapter interface used by the research agent

---

Quickstart (developer)

# clone
git clone <repo-url>
cd go-career-path

# install
npm install

# env
cp .env.example .env
# set LLM_PROVIDER, LLM_API_KEY, TAVILY_API_KEY, MCP_ENDPOINT, NEXT_PUBLIC_VERCEL_URL, etc.

# dev
npm run dev        # Next.js dev server

# build + start
npm run build
npm run start

---

Configuration (important keys)

• LLM_PROVIDER — provider id used by init_chat_model()
• LLM_API_KEY — API key for chosen provider
• TAVILY_API_KEY — Tavily search API key (if using default search)
• MCP_ENDPOINT — optional MCP server endpoint for enterprise search
• NEXT_PUBLIC_SITE_URL — Vercel/site URL for callbacks
• RESEARCH_AGENT_CONFIG — agent-specific timeouts, tool settings, and output schema

Ensure chosen models support:

• Structured output (JSON schema output)
• Tool calling (if using tool-enabled reasoning)

---

Usage (end-user)

• Via LangGraph Studio UI: configure model, run the research workflow, view the final report.
• Via web app: Input background → run agent → download or view final report with role suggestions and stepwise resources.

---

Output schema (example)

{
  "report_version": "1.0",
  "candidate_roles": [
    {
      "title": "Role A",
      "monthly_pay_estimate": "range",
      "why_it_fits": "rationale",
      "skills_to_build": ["skill1","skill2"],
      "learning_path": ["link1","link2"],
      "first_step": "actionable step"
    }
  ],
  "confidence_scores": { "Role A": 0.87 },
  "sources": ["url1","url2"],
  "timestamp": "2025-11-03T00:00:00Z"
}

Agents should output JSON matching your configured schema.

---

Best practices & tips

• Use stronger research models (e.g. gpt-4.1) for browsing and tool-enabled passes.
• Constrain token budgets for summarization/compression to keep costs predictable.
• Provide the agent with explicit role constraints (seniority, remote/on-site, industry) for targeted outputs.
• Validate final resource links programmatically or via an automated checker before presenting as canonical guidance.

---

Roadmap

• Fine-tune recommended learning tracks by industry and seniority.
• Add resume/CV tailoring generator tied to selected pivot role.
• Integrate verified course catalog & micro-credential connectors (Coursera, edX, LinkedIn Learning).
• Expand MCP adapters for enterprise/private knowledge graphs.

---

Contributing

• Fork, branch, and open PRs for feature work.
• Implement new search adapters under /packages/search-adapters.
• Add model provider connectors via init_chat_model() extensions.
• Follow project linting, testing, and PR guidelines in CONTRIBUTING.md.

---

License

MIT

---