A local-first, production-grade career intelligence engine that runs 24/7 on your machine — scraping opportunities, scoring them against your real skill profile, generating tailored resumes, and delivering a morning briefing to your phone. No cloud dependency. No vendor lock-in.

Most job search tools are passive — they wait for you to show up. ARIA runs whether you're at your desk or not.

Every morning, a 7-stage orchestrated pipeline runs automatically:

1. Scrapes live job postings from Built In Boston and MongoDB-cached sources
2. Scores every posting against your actual skill profile (55 verified skills) using sentence-transformers all-MiniLM-L6-v2 — runs fully local, no API calls
3. Qualifies jobs above threshold into a ranked pipeline
4. Generates archetype-matched resumes (AI_ML, Data_Engineering, Analytics, Governance_Ethics) via Ollama/Mistral (local LLM)
5. Syncs results into Memory Brain (SQLite) for cross-session intelligence
6. Composes a daily HTML digest with qualified jobs and application status
7. Delivers morning briefing to iPhone via iMessage — no app required

The dashboard at localhost:5100 gives you a live view of your pipeline, job scoring, and the embedded JAIA Resume Builder for on-demand resume generation against any job description.

🖱️ View Interactive Version → Click any node to explore technical details, stack, and implementation notes.

Design principle: local-first. All AI generation routes through Ollama. No cloud API costs, no data leaving the machine, no rate limits.

ARIA scores each job posting against 55 verified skills loaded from rosalina_resume_master.json, using sentence-transformers all-MiniLM-L6-v2. No cloud API. No Ollama. Runs entirely on-device.

# _score_job() in aria_integrations.py
job_emb    = self._semantic_matcher.model.encode([job_text])
skills_emb = self._semantic_matcher.model.encode(self._semantic_matcher.all_skills)
sims       = cosine_similarity(job_emb, skills_emb)[0]

# Average of top 5 skill matches — rewards depth, not breadth
semantic_score = float(np.mean(np.sort(sims)[-5:]))

# Composite score
confidence = (
    semantic_score * 0.65 +   # Primary signal
    location_score * 0.20 +   # Boston / remote match
    recency_score  * 0.15     # Posted within 14 days
) - neg_penalty               # Deducted for negative keyword hits

Jobs with excluded companies (configurable in aria_job_config.py) are zeroed out before scoring. Keyword matches are stored separately on the job document for display in the dashboard.

Embedded in the ARIA dashboard as a dedicated tab, the JAIA (Job Application Intelligence Agent) Resume Builder provides:

• Left panel: Paste any job description → AI extracts requirements → Ollama/Mistral enhances bullet points against your real project history
• Right panel: 6-section resume builder (Summary, Experience, Projects, Skills, Education, Certifications)
• Export: One-click .docx download

Backend lives at ~/Documents/memory-brain-integration/resume_server.py (port 5200), separate from the main repo. All generation is local via Ollama — no OpenAI key required.

ARIA is one node in a larger autonomous career operating system. The 7-stage rose_orchestrator.py coordinates:

Stage 1  aria_scrape         → Scrapes and caches live jobs
Stage 2  bridge              → Normalizes ARIA jobs for downstream agents
Stage 3  pod_pipeline        → Scout → Analyst → Ghostwriter agent chain
Stage 4  canvas_sync         → NEU Canvas → Notion (academic deadlines)
Stage 5  health_check        → Validates Memory Brain + pipeline integrity
Stage 6  digest              → Generates HTML daily summary
Stage 7  imessage_briefing   → Delivers morning briefing to iPhone

Stage isolation means a failure in Canvas sync doesn't block the job pipeline or iMessage delivery. Each stage logs to central_memory.db via the Memory Brain Data Bus.

• Python 3.11+
• MongoDB running locally (mongod --config /opt/homebrew/etc/mongod.conf --fork)
• Ollama with mistral and llama3 pulled (ollama pull mistral)
• Secrets configured at ~/.aria_secrets

# Start everything
bash ~/Desktop/launch_career.sh

# Or manually:
mongod --config /opt/homebrew/etc/mongod.conf --fork
cd ~/Desktop/Projects/aria-career-assistant
source venv/bin/activate
python aria_api.py
# Dashboard → http://localhost:5100
# Resume server → http://localhost:5200

cp .env.example .env
# Configure: MongoDB URI, Gmail IMAP credentials, Ollama endpoint

aria-career-assistant/
├── aria_api.py                  # FastAPI backend + dashboard server (port 5100)
├── aria_dashboard.html          # Frontend dashboard with Resume Builder tab
├── aria_integrations.py         # OpportunityScanner, ApplicationTracker, alerts
├── aria_job_config.py           # Target titles, keywords, scoring weights
├── resume_builder.py            # Resume builder script (in-repo)
│                                # Note: resume_server.py (port 5200) lives at
│                                # ~/Documents/memory-brain-integration/resume_server.py
├── aria_diagram.svg             # Animated architecture diagram (inline README)
├── aria_architecture.html       # Interactive architecture diagram (GitHub Pages)
├── src/
│   └── aria.py                  # Core agent entry point
├── config/
│   └── aria_config.yaml         # Endpoints, thresholds, archetype weights
├── requirements.txt
├── .env.example
└── .gitignore

• Semantic job scoring (all-MiniLM-L6-v2, 55-skill profile)
• MongoDB pipeline with 147+ job records
• FastAPI backend with live dashboard
• Archetype detection (4 career modes)
• Local LLM resume generation via Ollama
• ROSE OS 7-stage orchestration with stage isolation
• iMessage morning briefing
• JAIA Resume Builder embedded in dashboard (port 5200)
• Memory Brain integration (central_memory.db)

• LinkedIn outbound connection automation
• Recruiter email drafting agent
• GitHub repo sync from resume projects
• Application status auto-tracking from Gmail

• Multi-user support (separate MongoDB namespaces)
• REST API for external integrations
• Weekly analytics digest with trend analysis
• A/B testing on resume archetype performance

Program: MS Data Analytics Engineering — Northeastern University (EDGE) Expected Graduation: August 2026 Role: Graduate Research Associate & Conference Presenter

ARIA is a production system, not a class project. It runs daily and directly supports an active job search targeting ML/AI engineering roles. The ROSE OS orchestration layer was designed and built concurrently with coursework in Statistical Learning (IE7600) and Machine Learning & Analytics (DADS7275).

Rosalina Torres — ML/AI Engineer MS Data Analytics Engineering @ Northeastern University

• LinkedIn: linkedin.com/in/rosalina-torres
• GitHub: @rosalinatorres888
• Portfolio: rosalina.sites.northeastern.edu
• Email: torres.ros@northeastern.edu

MIT License — See LICENSE file for details

Part of the ROSE OS career intelligence ecosystem — built to replace manual job searching with autonomous, data-driven opportunity detection.