PASA - Personal Autonomous Software Assistant (MVP)

Setup (local dev)

git clone <repo-url>
cd pasa
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

Environment variables

Set these before running. You can export in your shell or place them in a .env file.

# Database
DATABASE_URL="postgresql+psycopg2://user:pass@host/db"

# Email (SMTP)
EMAIL_ADDRESS="[email protected]"
 EMAIL_PASSWORD="your-app-password"   # use an app password
 SMTP_SERVER="smtp.gmail.com"
 SMTP_PORT="587"

# User profile (stored in DB and used for applications)
 USER_ID="john"                    # any stable identifier
 USER_NAME="john doe"
 USER_EMAIL="[email protected]"
 USER_GITHUB=""      # optional
 USER_LINKEDIN="" # optional
 USER_TWITTER=""                          # optional
 USER_PROFILE_CONTEXT="- 5y Python/LLMs\n- Built data pipelines (example)"  # short paragraph or bullets
 USER_RESUME_PATH="/absolute/path/to/resume.pdf"  # e.g., /Users/you/path/PASA/data/resume/Your_Resume.pdf

# LLM (Groq OpenAI-compatible API)
 GROQ_API_KEY="your_groq_key"
 GROQ_BASE_URL="https://api.groq.com/openai/v1"   # default
 GROQ_MODEL="llama-3.3-70b-versatile"

# Data source
 WHATSAPP_EXPORT_PATH="data/whatsapp/group.txt"

Database migrations

Use Alembic to apply schema updates:

alembic upgrade head

Notes:

• Creating new tables is handled at runtime, but altering existing tables requires migrations.
• If you added new columns and see errors like UndefinedColumn, ensure you ran alembic upgrade head.

Run the pipeline

python -m src.orchestrator

What happens:

• User profile is upserted from env vars into the DB (including reading and storing resume bytes, filename, and mime type if USER_RESUME_PATH is set).
• WhatsApp export is parsed, potential job posts are detected, and posts are stored.
• Optional enrichment fetches job page text and computes a simple relevance score to your profile context.
• For posts with emails, an LLM (Groq) generates a tailored subject/body and an email is sent with your stored resume attached.
• If a Google Form is present (experimental), it attempts to auto-fill.

MVP

The current MVP includes:

• WhatsApp ingestion: parse data/whatsapp/group.txt into structured messages.
• NLP job detector: identify likely job posts and extract links and emails.
• Persistence: store job_posts with timestamp, sender, message, links, emails.
• Enrichment: fetch linked page text and compute relevance score to profile context.
• User profile management: upsert a single user_profile from environment variables; stores profile context and resume bytes for attachment.
• Email actions:
  • Tailored email generation via Groq (OpenAI-compatible) using GROQ_API_KEY and GROQ_MODEL.
  • SMTP send with optional resume attachment pulled from the DB.
• Form filler (experimental): basic Google Form auto-fill for simple field maps.

Troubleshooting

• Resume not attached:
  • Ensure USER_RESUME_PATH is an absolute path and the file exists.
  • Run alembic upgrade head so user_profile has resume_* columns.
  • After changing USER_RESUME_PATH, rerun the pipeline to refresh the stored resume.
• LLM not generating emails:
  • Verify GROQ_API_KEY is set; optionally set GROQ_BASE_URL and GROQ_MODEL.
  • Confirm openai package is installed (pip install -r requirements.txt).
• DB column errors:
  • Run alembic upgrade head to apply migrations.