lead-gen

Job lead pipeline — Chrome extension captures offers while browsing, FastAPI backend scores them with Mistral AI, and a Streamlit CRM dashboard tracks the full application funnel.

---

How it works

Browse job board
  → Chrome extension extracts title / company / description
  → POST /api/ingest (FastAPI)
  → Mistral AI scores relevance 0–10 + structured fields
      (matched skills, missing skills, mission types, seniority fit, nice-to-have match)
  → appended to data/jobs.csv
  → Streamlit CRM dashboard for review, status updates, outreach tracking

---

Stack

Layer	Tool
Browser capture	Chrome extension (MV3) — LinkedIn, APEC, WTTJ, Indeed, HelloWork
Backend API	FastAPI + Uvicorn
Scoring	Mistral AI (mistral-small-latest)
Storage	CSV (data/jobs.csv)
CRM dashboard	Streamlit

---

Local setup

1. Backend

cd backend
python -m venv .venv
.venv\Scripts\activate        # Windows
pip install -r requirements.txt

# create .env from template
cp .env.example .env
# fill in MISTRAL_API_KEY

uvicorn app:app --reload
# → http://127.0.0.1:8000

2. Streamlit CRM

# from repo root (with .venv active)
streamlit run streamlit/app.py
# → http://localhost:8501

3. Chrome extension

1. Open chrome://extensions
2. Enable Developer mode
3. Click Load unpacked → select the extension/ folder

The extension injects a capture button on LinkedIn, APEC, WTTJ, Indeed, and HelloWork job pages.

---

Candidate persona

Scoring is driven by backend/persona.md — a plain-text file with two sections:

• Profile — experience level, strong/weak skills, target cities, contract type
• Projects — short descriptions of built projects used to calibrate seniority_fit and nice_to_have_match

Edit persona.md directly to update how jobs are scored. No code changes needed.

---

Backfill / rescore

When the scoring prompt changes or new fields are added, existing CSV rows can be updated without re-browsing the job boards:

# rescore only rows with empty matched_skills (stale rows)
python scripts/backfill_scores.py

# rescore every row
python scripts/backfill_scores.py --force

Saves progress after each row — safe to interrupt.

---

Scoring output fields

Each ingested job is enriched with:

Field	Type	Description
relevance_score	int 0–10	Overall relevance. Penalizes missing required skills.
fit_explanation	str	One-sentence summary of the score
mission_types	list	Role focus: pipeline, transformation, cloud_migration, analytics_engineering, data_platform, consulting
required_stack	list	Tools explicitly required by the posting
bonus_stack	list	Tools listed as nice-to-have
matched_skills	list	Required skills the candidate has
missing_skills	list	Required skills the candidate is missing
nice_to_have_match	list	Bonus criteria the candidate matches
seniority_fit	int	0 = in range (0–3y), 1 = stretch (3–5y), 2 = out of league (5y+)

---

Project structure

backend/
  app.py          FastAPI routes (ingest, list, patch, delete)
  models.py       Pydantic models (JobPosting, JobRecord, JobUpdate)
  scorer.py       Mistral AI scoring logic
  crm.py          CSV read/write/delete helpers
  persona.md      Candidate profile — edit to update scoring context
extension/
  manifest.json   Chrome MV3 manifest
  background.js   Service worker
  content/
    linkedin.js   LinkedIn job page scraper
    apec.js       APEC job page scraper
    wttj.js       WTTJ job page scraper
    indeed.js     Indeed job page scraper
    hellowork.js  HelloWork job page scraper
  popup/
    popup.html    Extension popup UI
    popup.js      Popup logic
    popup.css
scripts/
  backfill_scores.py  Rescore stale or all CSV rows from the terminal
streamlit/
  app.py          CRM dashboard
data/
  jobs.csv        Local job store (git-ignored in production)