{"repo": "ayhm23/resume-env", "files": {"README.md": {"c6060376e72d44bb73ab525b651a2d96de95c588": {"t": 1785846879.0244856, "n": 6727, "h": "a1b082443e95bab3", "c": "---\ntitle: resume-env\nemoji: 📄\ncolorFrom: blue\ncolorTo: green\nsdk: docker\napp_port: 8000\ntags:\n  - rl-environment\n  - resume\n  - ats\n  - job-application\nlicense: bsd-3-clause\n---\n\n# ResumeEnv 🎯\n**Job Application Optimizer — OpenEnv Environment**\n\nAn OpenEnv-compliant environment where an AI agent reads a job description and a candidate resume, then optimizes the resume and cover letter to maximize ATS (Applicant Tracking System) match rate.\n\n---\n\n## Install & Run\n\n```bash\n# Install client\npip install git+https://huggingface.co/spaces/ayhm23/resume-env\n\n# Run locally with Docker\ndocker build -t resume-env .\ndocker run -p 8000:8000 resume-env\n```\n\n## Quick Start\n\n```python\nimport asyncio\nfrom resume_env import ResumeEnv, ResumeAction\n\nasync def main():\n    async with ResumeEnv(base_url=\"https://ayhm23-resume-env.hf.space\") as env:\n\n        # Task 1 — Keyword Extraction\n        result = await env.reset(task_id=\"task1_keyword_extraction\")\n        print(result.observation.job_description)\n\n        result = await env.step(ResumeAction(\n            action_type=\"extract_keywords\",\n            hard_skills=[\"SQL\", \"Python\", \"Tableau\"],\n            soft_skills=[\"stakeholder communication\"],\n            experience_years=5,\n        ))\n        print(f\"Score: {result.reward}\")  # e.g. 0.78\n\nasyncio.run(main())\n\n# Sync usage\nfrom resume_env import ResumeEnv, ResumeAction\nwith ResumeEnv(base_url=\"http://localhost:8000\").sync() as env:\n    result = env.reset(task_id=\"task2_bullet_rewrite\")\n    result = env.step(ResumeAction(\n        action_type=\"rewrite_bullet\",\n        rewritten_bullet=\"Developed Tableau dashboards tracking 12 KPIs...\",\n    ))\n    print(result.reward)\n```\n\n---\n\n## API Endpoints\n\n| Endpoint    | Method | Description |\n|-------------|--------|-------------|\n| `/reset`    | POST   | Start new episode. Body: `{\"task_id\": \"task1_keyword_extraction\"}` |\n| `/step`     | POST   | Submit action, get observation + reward |\n| `/state`    | GET    | Episode metadata (step_count, episode_id, cumulative_reward) |\n| `/tasks`    | GET    | List all tasks with action schemas |\n| `/grader`   | GET    | Last episode grader scores |\n| `/baseline` | POST   | Run rule-based baseline on all 3 tasks |\n| `/health`   | GET    | `{\"status\": \"ok\"}` |\n| `/ws`       | WS     | WebSocket (used by Python client) |\n| `/web`      | GET    | Interactive browser UI |\n\n---\n\n## Tasks\n\n### Task 1 — Keyword Extraction (Easy) · max_steps=1\n\n**Goal:** Extract structured skills from a job description.\n\n**Action:**\n```json\n{\n  \"action_type\": \"extract_keywords\",\n  \"hard_skills\": [\"SQL\", \"Python\", \"Tableau\"],\n  \"soft_skills\": [\"stakeholder communication\"],\n  \"experience_years\": 5\n}\n```\n\n**Reward:** `0.60 × hard_skill_F1 + 0.25 × soft_skill_Jaccard + 0.15 × experience_accuracy`\n\n---\n\n### Task 2 — Bullet Rewrite (Medium) · max_steps=1\n\n**Goal:** Rewrite one resume bullet for maximum ATS match.\n\n**Action:**\n```json\n{\n  \"action_type\": \"rewrite_bullet\",\n  \"rewritten_bullet\": \"Developed Tableau dashboards tracking 12 KPIs, reducing reporting time by 40%.\"\n}\n```\n\n**Reward:** `0.40 × ATS_improvement + 0.30 × semantic_preservation + 0.30 × quality`\n\nQuality = strong action verb (0.5) + quantification present (0.5).\n\n---\n\n### Task 3 — Full Application Pack (Hard) · max_steps=4\n\n**Goal:** Build a complete tailored application in 4 sequential steps.\n\n| Step | action_type | content |\n|------|-------------|---------|\n| 1 | `rewrite_summary` | New 2-sentence professional summary |\n| 2 | `rewrite_experience` | Rewritten bullets, one per line |\n| 3 | `update_skills` | Comma-separated updated skills list |\n| 4 | `write_cover_letter` | Full cover letter starting with \"Dear Hiring Manager,\" |\n\n**Final Reward:** `0.35 × resume_ATS + 0.25 × cover_relevance + 0.25 × keyword_coverage + 0.15 × format_score`\n\nPartial rewards are given after each step.\n\n---\n\n## Observation Space\n\nEvery observation (`ResumeObservation`) contains:\n\n| Field | Type | Description |\n|-------|------|-------------|\n| `task_id` | str | Current task |\n| `job_description` | str | Full JD text |\n| `resume_snapshot` | dict | Current resume state |\n| `feedback` | str | Human-readable score breakdown |\n| `current_score` | float | Score so far |\n| `steps_remaining` | int | Steps left in episode |\n| `done` | bool | Episode complete |\n| `reward` | float | Step reward |\n| `metadata` | dict | Detailed grader breakdown |\n\n---\n\n## Environment Variables\n\n`inference.py` reads these from the shell environment (or a `.env` file):\n\n| Variable | Required | Description | Example |\n|----------|----------|-------------|---------|\n| `HF_TOKEN` | **Yes** | Your API key (HF or Groq or OpenAI) | `hf_...` / `gsk_...` / `sk-...` |\n| `API_BASE_URL` | No | LLM provider endpoint. **If unset**, the OpenAI SDK defaults to `https://api.openai.com/v1` | `https://api.groq.com/openai/v1` |\n| `MODEL_NAME` | No | Model identifier for the chosen provider. Default: `gpt-4o-mini` | `llama-3.3-70b-versatile` |\n| `ENV_BASE_URL` | No | Running ResumeEnv server. Default: `http://localhost:8000` | `https://ayhm23-resume-env.hf.space` |\n\n> **Note:** `OPENAI_API_KEY` is also accepted as an alias for `HF_TOKEN`.\n\n### Using Groq\n```bash\nexport HF_TOKEN=gsk_...\nexport API_BASE_URL=https://api.groq.com/openai/v1\nexport MODEL_NAME=llama-3.3-70b-versatile\npython inference.py\n```\n\n### Using OpenAI\n```bash\nexport HF_TOKEN=sk-...\n# API_BASE_URL not needed — SDK defaults to api.openai.com\nexport MODEL_NAME=gpt-4o-mini\npython inference.py\n```\n\n---\n\n## Baseline\n\n```bash\n# Rule-based (no LLM, always works)\ncurl -X POST http://localhost:8000/baseline\n\n# LLM baseline\nexport OPENAI_API_KEY=sk-...\nexport MODEL_NAME=gpt-4o-mini\npython inference.py\n```\n\nExpected baseline scores:\n\n| Task | Rule-based | LLM (gpt-4o-mini) |\n|------|-----------|-------------------|\n| Task 1 | ~0.48 | ~0.80 |\n| Task 2 | ~0.55 | ~0.72 |\n| Task 3 | ~0.38 | ~0.65 |\n\n---\n\n## Project Structure\n\n```\nresume_env/\n├── __init__.py               ← exports ResumeAction, ResumeObservation, ResumeState, ResumeEnv\n├── models.py                 ← Action / Observation / State dataclasses\n├── client.py                 ← ResumeEnv(HTTPEnvClient)\n├── inference.py              ← LLM baseline agent (OpenAI-compatible)\n├── openenv.yaml              ← OpenEnv manifest\n├── pyproject.toml\n├── .dockerignore\n├── README.md\n└── server/\n    ├── __init__.py\n    ├── resume_environment.py ← ResumeEnvironment (reset / step / state)\n    ├── app.py                ← FastAPI (create_web_interface_app + /tasks /grader /baseline)\n    ├── requirements.txt      ← server deps (no ML libraries)\n    └── Dockerfile\n```\n\n---\n\n## Deploy\n\n```bash\ncd resume_env\nopenenv push --repo-id ayhm23/resume-env\n```\n\n---\n\n## License\nBSD-3-Clause\n"}}}}