Applination

AI-assisted job application pipeline (internships, new-grad, and full-time). Every day it fetches fresh job postings, uses an LLM to rank them against your background (0–100 fit score), tailors your resume, and writes a cover letter for each top match — all from a web dashboard or the CLI.

GitHub: https://github.com/sanaro99/applination

---

What it does

1. Fetches fresh job postings from 8 public APIs / aggregators
2. Uses an LLM to score each job against your background (0–100 fit)
3. Tailors your resume per job and writes a personalised cover letter
4. Enforces a one-page ATS-safe .docx / .pdf output
5. Produces a daily Excel tracker with status dropdowns and hyperlinks

---

Feature overview

Area	What you get
Daily pipeline run	Full fetch → rank → tailor loop; dry-run mode for free preview
Web dashboard	Live SSE progress stream, job cards with scores, cancel mid-run
Application tracker	Table + Kanban view; status, notes, deadline, tags, bulk edits, CSV export
Application detail	PDF preview, inline resume tweak (versioned), cover-letter editor, resume diff
Single-job wizard	Paste a URL or JD → AI extracts fields → review → generate tailored docs in 3 steps
Ranked triage	Full scored pool per run; rescue any non-auto-selected job on demand
Prepwork (Coach)	Chat with an AI coach grounded on your resume and stories; mock interview mode; essay drafter; reusable answer bank
Insights	Run history, log viewer, run compare (diff two runs), stats charts (scores, sources, status)
Setup	In-app config.yaml editor with live provider test, visual per-workflow LLM routing editor, master-data editor with AI-assist
Close the loop	Gmail inbox sync (IMAP) classifies recruiter emails → auto-updates application status; iCal feed for deadlines + interviews; daily email digest

---

Prerequisites

• Python 3.11+
• Node.js 20+ and npm
• At least one LLM provider key (or a running Ollama instance — free)
• PDF conversion (optional): Microsoft Word (Windows/macOS) or LibreOffice (soffice on PATH, Linux). Skip with --no-pdf if neither is available.

---

Getting started (recommended path)

1. Clone and install

git clone https://github.com/sanaro99/applination.git
cd applination

python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

pip install -r requirements.txt

cd web
npm install
cd ..

2. Start the development servers

.\scripts\dev.ps1

This starts:

• FastAPI backend on http://127.0.0.1:8000
• Next.js frontend on http://127.0.0.1:3000

Open http://localhost:3000 in your browser.

3. Complete onboarding

On first launch you'll be redirected to the onboarding wizard:

1. Provider — enter your LLM API key (Claude, Gemini, DeepSeek, etc.) or choose Ollama
2. Contact — name, email, phone, LinkedIn URL (used in rendered documents)
3. Resume — upload a PDF/DOCX or paste text; AI extracts it into structured master data
4. Interview — a short AI conversation fills experience gaps and drafts your story bank
5. Search prefs — job titles, locations, minimum match score, max jobs per day

Your data stays local in gitignored files (config.yaml, master_data/).

4. Run dry run (free — no tailoring LLM calls)

Click Start run → Dry run in the dashboard, or:

python -m src.main --dry-run

Fetches and ranks jobs; writes output/YYYY-MM-DD/ranked_jobs.json. No resume generation.

5. Full run

Click Start run in the dashboard, or:

python -m src.main

---

CLI quick reference

# Dry run
python -m src.main --dry-run

# Full run
python -m src.main

# Skip PDF (produce .docx only)
python -m src.main --no-pdf

# Tweak a generated resume
python -m src.tweak output/2026-04-24/Company_Role/resume.docx "Emphasize LangGraph work"
python -m src.tweak resume.docx "more ML focus" --provider gemini
python -m src.tweak resume.docx --interactive

---

Output artifacts

Each run creates output/YYYY-MM-DD/<Company_Role>/ containing:

File	Description
resume.docx / resume.pdf	Tailored one-page resume
resume.json	Structured JSON (input for tweak.py)
cover_letter.docx / cover_letter.pdf	Personalised cover letter
job.json	JD snapshot (company, title, description, URL)
answers.md	AI-drafted screening question answers (if any)

Plus output/YYYY-MM-DD/apps_YYYY-MM-DD.xlsx — daily Excel tracker.

---

Daily scheduling

macOS / Linux (cron)

bash scripts/setup_cron.sh

Windows (Task Scheduler)

.\scripts\setup_task_scheduler.ps1 -Time "08:00"

---

LLM providers

All 7 providers are supported; any can be primary or fallback. Configure in config.yaml or via the Setup → Workflows page.

Provider	Key env var	Notes
Claude (Anthropic)	ANTHROPIC_API_KEY	Haiku ~$0.10–0.30/run; Sonnet ~$1–3/run
Gemini (Google)	GOOGLE_API_KEY	Flash is effectively free
DeepSeek	DEEPSEEK_API_KEY	deepseek-chat for standard; deepseek-v4-pro for premium tier
Mistral	MISTRAL_API_KEY	Good mid-tier option
OpenRouter	OPENROUTER_API_KEY	Access to many models via one key
Ollama (local)	none	Fully free; requires ollama serve
Nvidia NIM	NIM_API_KEY	Self-hosted or cloud inference

Per-workflow routing (ranking, tailoring, cover letter, Coach, etc.) is configurable independently via the Workflows page.

---

Master data

File	Purpose
master_data/resume.yaml	Full master resume — source of truth for all content
master_data/bio.md	Voice/tone reference injected into cover-letter prompts
master_data/stories/*.md	Narrative stories with YAML frontmatter; auto-matched to jobs
master_data/cover_letters/examples/	Past letters as .md files; used as style anchors
master_data/guidelines/*.md	Resume/cover-letter writing rules matched per role

Stories use YAML frontmatter (tags, role_fit, company_fit, one_liner). The cover letter builder picks the 1–2 most relevant stories per job by tag overlap + keyword matching. Stories can be generated from the Master Data page using AI.

---

Onboarding flow detail

Step	What happens
Provider	Validates the key with a real test call; saved to config.yaml
Contact	Name/email/phone/LinkedIn written to config.yaml user: section
Resume import	PDF/DOCX/text → src/content_studio.import_resume() → master_data/resume.yaml
AI interview	Multi-turn conversation extracts experience gaps, drafts bio.md + initial stories
Search prefs	Keywords, locations, min_match_score, max_jobs_per_day saved to config.yaml search:

After onboarding, GET /api/onboarding/status returns can_run: true and the dashboard unlocks.

---

Configuration reference

Key sections in config.yaml:

user:
  name: "..."
  email: "..."
  phone: "..."
  linkedin: "..."

search:
  keywords: [software engineer intern, ...]
  locations: [Remote, New York, ...]
  min_match_score: 55       # 0-100 threshold; jobs below this are skipped
  max_jobs_per_day: 50      # cap on auto-tailored applications per run

llm:
  primary: "claude"
  fallbacks: ["gemini"]
  claude:
    api_key: ""
    model: "claude-haiku-4-5-20251001"
  # Per-workflow overrides:
  tasks:
    tailoring_premium:
      primary: "deepseek"
    coach:
      primary: "gemini"

output:
  root: "output"
  produce_pdf: true
  base_font_size: 10

inbox:                        # Close-the-loop inbox sync (optional)
  email: ""
  app_password: ""            # Gmail App Password (not your login password)
  auto_update_status: true
  min_confidence: 0.7

reminders:
  digest_enabled: false
  deadline_window_days: 3
  follow_up_days: 7

---

Known constraints

• Single-tenant: one profile per install (but any person's profile — not hardcoded to any one user)
• PDF conversion: requires MS Word (Windows/macOS) or LibreOffice (Linux); use --no-pdf otherwise
• LinkedIn scraping: not supported (rate-limited / requires login)
• No streaming in Coach: responses are send-and-wait; streaming not yet implemented
• Greenhouse slugs: the default list in config.example.yaml is illustrative — replace with companies you want to track