fu7ur3pr00f 0.1.2

Career intelligence system - gather professional data, generate CVs, analyze career alignment

FutureProof

Career intelligence agent that gathers professional data, searches job boards, analyzes career trajectories, and generates ATS-optimized CVs — all through conversational chat. Built with LangChain, LangGraph, and ChromaDB. Supports OpenAI, Anthropic, Google, Azure, and Ollama.

What It Does

You:   Gather all my career data
Agent: [gathers LinkedIn, portfolio, CliftonStrengths → indexes to ChromaDB]

You:   Analyze my skill gaps for Staff Engineer
Agent: [runs skill gap analysis using your data + market trends]

You:   Search for remote Python developer jobs in Europe
Agent: [queries 7 job boards + Hacker News hiring threads]

You:   Generate my CV targeting that Staff Engineer role
Agent: [generates ATS-optimized CV in Markdown + PDF]

One agent, 40 tools, 12 MCP clients. Data sources: LinkedIn CSV export, GitHub (live MCP), GitLab (glab CLI), portfolio websites, CliftonStrengths PDF, 7 job boards, Hacker News, Dev.to, Stack Overflow, Tavily search.

Architecture

graph LR
    User <-->|Rich UI, HITL| Chat[Chat Client]
    Chat <--> Agent[Single Agent<br/>40 tools]

    Agent --> Gather[Gatherers]
    Agent --> MCP[12 MCP Clients]
    Agent --> Analysis[Career Analysis]
    Agent --> Gen[CV Generator]

    Gather -->|LinkedIn CSV, Portfolio,<br/>CliftonStrengths| ChromaDB[(ChromaDB)]
    MCP -->|GitHub, 7 job boards,<br/>HN, Tavily, Dev.to, SO| Agent
    Analysis --> LLM[Multi-Provider LLM<br/>Fallback Chain]
    Gen -->|Markdown + PDF| Output[CV Output]

    ChromaDB -->|RAG search| Agent
    ChromaDB -->|Episodic memory| Agent
    LLM -->|Purpose-based routing| Agent

Key design decisions:

• Single agent — multi-agent handoffs failed with GPT-4.1 (over-delegation, lost context). One agent with all tools is simpler and more reliable.
• Database-first pipeline — gatherers return Section NamedTuples and index directly to ChromaDB. No intermediate files, no markdown header roundtrip.
• Two-pass synthesis — GPT-4o genericizes analysis responses regardless of prompt engineering. AnalysisSynthesisMiddleware lets the agent do tool calling, then replaces its generic response with focused synthesis from a reasoning model.
• Multi-provider fallback — supports OpenAI, Anthropic, Google, Azure, Ollama, and FutureProof proxy. Provider-specific fallback chains with automatic rate-limit recovery and purpose-based routing (agent/analysis/summary/synthesis).
• HITL confirmation — destructive or expensive operations (CV generation, full data gathering, knowledge clearing) require user approval via LangGraph's interrupt().

Quick Start

pipx install fu7ur3pr00f
fu7ur3pr00f

If fu7ur3pr00f is not found, run pipx ensurepath and restart your shell.

On first launch, the /setup wizard prompts you to configure an LLM provider. Supports OpenAI, Anthropic, Google, Azure, Ollama, or the FutureProof proxy. Settings are saved to ~/.fu7ur3pr00f/.env. Everything happens inside the chat — use /help to see all commands.

Install via apt (Debian/Ubuntu, amd64)

curl -fsSL https://juanmanueldaza.github.io/fu7ur3pr00f/fu7ur3pr00f-archive-keyring.gpg | \
  sudo tee /usr/share/keyrings/fu7ur3pr00f-archive-keyring.gpg >/dev/null

echo "deb [signed-by=/usr/share/keyrings/fu7ur3pr00f-archive-keyring.gpg] \
https://juanmanueldaza.github.io/fu7ur3pr00f stable main" | \
  sudo tee /etc/apt/sources.list.d/fu7ur3pr00f.list >/dev/null

sudo apt update
sudo apt install fu7ur3pr00f

The apt package bundles github-mcp-server and installs glab, poppler-utils, and WeasyPrint system libraries as dependencies.

PDF generation (CVs) requires system libraries for text rendering. Without them the app works fine — you just get Markdown output instead of PDF.

Ubuntu/Debian: sudo apt-get install libpango-1.0-0 libpangoft2-1.0-0 libcairo2 libfontconfig1 libgdk-pixbuf-2.0-0 poppler-utils

macOS: brew install pango cairo gdk-pixbuf poppler

Project Structure

src/fu7ur3pr00f/
├── agents/
│   ├── career_agent.py     # Single agent: create_agent(), 4 middlewares, singleton cache
│   ├── middleware.py        # Dynamic prompt, synthesis, tool repair, summarization
│   ├── orchestrator.py      # LangGraph Functional API for analysis workflows
│   ├── helpers/             # Orchestrator support (data pipeline, LLM invoker)
│   └── tools/              # 40 tools by domain (profile, gathering, analysis, market, settings)
├── chat/                   # Streaming client, HITL interrupt loop, Rich UI, /setup wizard
├── gatherers/              # LinkedIn CSV, CliftonStrengths PDF, portfolio scraper, market data
├── generators/             # CV generation (Markdown + PDF via WeasyPrint)
├── llm/                    # FallbackLLMManager: multi-provider fallback, purpose-based routing
├── memory/                 # ChromaDB (knowledge RAG + episodic), chunker, profile, embeddings
├── mcp/                    # 12 MCP clients: GitHub, Tavily, financial, 7 job boards (incl. HN), Dev.to, SO
├── prompts/                # System + analysis + CV prompt templates
├── services/               # GathererService, AnalysisService, KnowledgeService
└── utils/                  # PII anonymization, data loading, logging

Development

git clone https://github.com/juanmanueldaza/fu7ur3pr00f.git
cd fu7ur3pr00f
pip install -e .
pip install pyright pytest ruff    # dev tools

pytest tests/ -q              # Unit tests
pyright src/fu7ur3pr00f       # Type checking
ruff check .                  # Lint

Fresh Install Connectivity Check

Use this to validate a clean pipx install plus MCP/LLM connectivity from a temporary HOME.

scripts/fresh_install_check.sh --source local --config-from .env
scripts/fresh_install_check.sh --source pypi --config-from .env

Tech Stack

Python 3.13 · LangChain + LangGraph · ChromaDB · Multi-provider LLM (OpenAI, Anthropic, Google, Azure, Ollama) · Typer + Rich · WeasyPrint · httpx

---

Licensed under GPL-2.0.