fu7ur3pr00f 0.1.0

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, 41 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/>41 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

pip install futureproof
futureproof

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 ~/.futureproof/.env. Everything happens inside the chat — use /help to see all commands.

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/futureproof/
├── 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/              # 41 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, 6 job boards, HN, financial, content
├── 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 futureproof
pip install -e .
pip install pyright pytest ruff    # dev tools

pytest tests/ -q              # Unit tests
pytest tests/eval/ -m eval    # Eval tests (need Azure credentials)
pyright src/futureproof       # Type checking
ruff check .                  # Lint

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.