The Last Application

⚠️ Work in Progress — under active development.

Overview

One bot to rule all applications, one bot to find them,
One bot to bring all offers and in employment bind them.

Automates LinkedIn Easy Apply end-to-end: finds jobs, answers application questions using a retrieval-augmented AI over your PDFs and personal notes, and logs everything in a local DuckDB. Designed for reliability, concise answers, and easy extension.

Features

• Antibot-aware automation
  • Connects to an existing Chromium CDP session for realistic control.
• Smart answering
  • Defaults first: data/defaultanswers.json by label.
  • RAG fallback: FAISS over data/*.pdf with data/myinfo.txt as authoritative context.
  • Strict output rules: numbers only, Yes/No, or concise text when required.
  • Guaranteed fallback: if AI fails or is disabled, uses deterministic answers.
• Persistence
  • DuckDB with jobs, hirers, sessions, applications, form questions and responses.
  • Bulk inserts, indices, stats, CSV import/export.
• Easy Apply flow
  • Inputs, selects, radios/checkboxes supported.
  • Iterates steps, handles inline feedback, submits, and records Q&A.
• Extensible
  • Clear module boundaries for scraping, AI, and storage.

Quick start

1. Launch Chromium using terminal in the base directory(must be running before the app)

chromium-browser --remote-debugging-port=9222 --user-data-dir=data/chromium_user_data --no-first-run --no-default-browser-check --no-sandbox --disable-gpu --disable-dev-shm-usage

2. Login to LinkedIn on the browser

3. Install Python deps (3.10+ recommended)

pip install -r requirements.txt

4. Prepare env and data

• Create .env with OPENAI_API_KEY=your_key_here
• Configure config/tla_config.yaml with:
• Populate data/:
  • defaultanswers.json — known label→answer map
  • myinfo.txt — short personal facts (authoritative)
  • *.pdf — resume/portfolio/certs for retrieval

5. Run

   python main.py
   

Usage notes

• Default → AI → fallback
  • If label found in defaultanswers.json, use it directly.
  • Else, if use_ai is true, call AI RAG over PDFs with myinfo.txt as authoritative context.
  • Else, or if AI returns nothing/invalid, use deterministic fallback (safe option, empty, or rule-based).
• Strict outputs
  • Numbers-only when asked for numeric/years.
  • Exactly Yes/No for binary.
  • Otherwise concise single word/short sentence.
• Storage
  • Jobs and applications, plus each question/response, are persisted in DuckDB under database/.

Configuration reference

• .env
  • OPENAI_API_KEY: required for current provider (ChatGPT path).
• config/nodriver_config.yaml
  • connection: host, port, userDataDir, headless, noSandbox.
  • search: baseUrl (LinkedIn jobs), query (keyword), geoId (location), easyApply (bool).
  • langchain: model_name, temperature, using_ollama (WIP; keep false for now), openai_api_key (optional).
• data/
  • defaultanswers.json: hard overrides by label.
  • myinfo.txt: authoritative add-on context (short and factual).
  • PDFs: indexed by FAISS for retrieval.

Architecture

• Scraper: opens LinkedIn search, iterates job cards, extracts details, and stores jobs.
• RAG: FAISS index over PDFs; myinfo.txt injected as authoritative additional_context.
• Chain: retriever → stuff-docs chain → LLM with strict prompt rules; post-processed if needed.
• Form runner: discovers fields, applies defaults/AI/fallback, advances steps, and submits.
• DB: DuckDB schema for jobs, hirers, sessions, applications, questions, responses; CSV IO and stats.

Checklist before running

• Chromium is launched with:
  • --remote-debugging-port=9222
  • --user-data-dir=data/chromium_user_data
  • --no-first-run --no-default-browser-check
  • --no-sandbox --disable-gpu --disable-dev-shm-usage
• OPENAI_API_KEY set in .env.
• config/nodriver_config.yaml has:
  • connection.host=127.0.0.1, port=9222, userDataDir=data/chromium_user_data
  • search fields (baseUrl, query, geoId, easyApply)
  • langchain.model_name set; using_ollama=false
• data/ contains:
  • defaultanswers.json
  • myinfo.txt
  • at least one PDF for retrieval (resume recommended)
• use_ai flag set as desired:
  • true: use AI with guaranteed fallback
  • false: always use deterministic fallback if not found in defaults
• Database/ directory is writable (created automatically if missing).

Roadmap

• Ollama support
  • Add parity path for local models; model selection and graceful provider fallback.
• Stronger fallback strategies
  • Per-field deterministic rules, option ranking heuristics, and inline-feedback driven retries.
• Stability hardening
  • Better waits/retries, selector updates, anti-fingerprint refinements.
• Dashboard
  • Visualization for throughput, success rates, and answer quality.

Disclaimer

Automating third-party websites may violate their terms. Use responsibly and at your own risk.