An AI-powered recruiting pipeline. It extracts resumes from new applications, has Llama 3.3 70B structure candidate profiles into strict JSON, scores each candidate against live job requirements with a weighted engine, logs every decision to a hiring pipeline, and routes outreach, Calendly scheduling for shortlists, polished declines for rejects, and a real-time Slack digest for the team.
Service
Automation
Industry
SaaS & Tech
Category
AI Automation
Built with: Groq / Llama, n8n, Google Sheets, Calendly.
Overview
The Problem
Manual resume screening is slow, inconsistent, and breaks down as application volume grows
Recruiters waste hours opening PDFs, comparing skills to job requirements, and rewriting the same accept/decline emails
Strong candidates wait too long while weak applications clog the pipeline
There's no structured record of fit score, missing skills, or why someone was shortlisted, held, or rejected
The Solution
New applications trigger the workflow automatically from Google Sheets, intake starts the moment a row is added
Resumes are downloaded from Google Drive, parsed to text, and shaped into a structured candidate object
Llama 3.3 70B extracts normalized skills and experience into strict JSON; a weighted engine scores fit against live requirements
Results are logged to a pipeline sheet, candidates are routed by score threshold, and recruiters get a Slack digest in real time
Solution Workflow
Trigger
Extract
Structure
Analyze
Score
Route
Full Technical Breakdown
1. Application Intake & Resume Extraction
A Google Sheets Trigger fires on rowAdded when a candidate submits an application. The workflow resolves the Google Drive file ID from the row, downloads the resume PDF, and extracts the full text, turning an unstructured upload into machine-readable input.
2. Candidate Object Construction
A Code node maps intake columns into a clean candidate object, Full Name, Email, Phone, Position, Years of Experience, Location, LinkedIn, and GitHub, then attaches the extracted resume text. This normalized payload becomes the single source of truth for AI analysis, scoring, emails, and logging.
3. AI ATS Extraction (Llama 3.3 70B)
A Basic LLM Chain running llama-3.3-70b-versatile via Groq acts as a technical recruiter/ATS, prompted to return only valid JSON covering summary, skills, languages, frameworks, databases, cloud platforms, tools, education, certifications, experience, projects, strengths, and weaknesses. Skills are normalized and missing fields return empty values instead of invented data. A Parse AI JSON node strips fence artifacts and merges contact fields for downstream use.
4. Weighted Scoring Engine
Candidate JSON is merged with live job requirements from Google Sheets, Required Skills, Preferred Skills, Minimum Experience, and Minimum Score. Custom logic fuzzy-matches skills across all stack categories, then computes an explainable fit score: 60% required skills, 20% preferred skills, 20% experience, with matched/missing skills and plain-text reasoning for every decision.
5. Threshold Routing & Branded Email Outreach
Score bands drive automated communication: at or above the minimum score routes to an HTML Gmail with a Calendly scheduling CTA, around 70% of the minimum holds for manual review, and anything below sends a polished HTML decline, both templates fully styled inline HTML, ready for inbox rendering.
6. Pipeline Logging & Slack Digest
Every scored candidate is appended to a Google Sheets hiring pipeline with fit score, status, and reasoning. A digest is built and sent to Slack so recruiters see shortlists and review cases immediately, without babysitting the spreadsheet.
Results
Key Features
Event-driven application intake: Screening starts the moment a new Google Sheets row is added, no manual batch runs, no forgotten applications sitting unread.
Resume-to-structure pipeline: Downloads the PDF from Google Drive, extracts text, and builds a clean candidate object from form fields plus resume content, unstructured applications become structured hiring data.
Strict JSON ATS extraction: Llama 3.3 70B extracts a fixed ATS schema with skill normalization and anti-hallucination rules, portfolio-grade structure, not free-form AI summaries.
Explainable weighted scoring: Transparent fit scoring, 60% required skills, 20% preferred, 20% experience, with fuzzy matching, missing-skill lists, and written reasoning for every outcome.
Smart threshold routing: A configurable minimum score drives three paths: Calendly invite for shortlists, manual review for the mid-band, and branded decline emails for clear rejects, speed where it's safe, humans where it matters.
Pipeline logging + Slack visibility: Every decision is logged to Sheets, and Slack keeps the hiring team informed in real time, full audit trail, zero spreadsheet babysitting.
Before
Recruiters open each PDF manually, compare skills by eye, rewrite accept/decline emails from scratch, lose track of why decisions were made, and strong candidates wait while volume piles up.
After
Applications trigger automatic resume extraction and Llama-powered ATS parsing, a weighted engine produces an explainable fit score, shortlisted candidates get a Calendly scheduling email, rejects get a polished decline, mid-band cases stay in manual review, and the team is alerted on Slack instantly.
What This Proves
This system shows that recruiting operations can combine document extraction, constrained LLM structuring, and deterministic scoring into one reliable pipeline, automating first-pass screening while keeping humans in the loop for borderline decisions.
Gmail API — Shortlist and decline HTML email delivery
Calendly — 30-minute screening-call scheduling
Slack API — Recruiter digest notifications
HTML / CSS — Branded candidate-facing email templates
Scalability & Flexibility
LLM Provider Flexibility: Currently runs on Groq Llama 3.3 70B; can swap to GPT-4o, Claude, or Gemini for deeper resume reasoning or multilingual screening.
Config-Driven Roles: Required Skills, Preferred Skills, Minimum Experience, and Minimum Score live in Sheets, new roles are screened by updating data, not rewriting the graph.
Intake Source Expansion: Trigger can extend from Google Sheets to Typeform, Greenhouse, Lever, Ashby, or a careers-page webhook without redesigning scoring or routing.
OCR for Scanned Resumes: Text extraction can add OCR for image-based PDFs so non-selectable resumes don't drop out of the pipeline.
Human-in-the-Loop Review: Screening-band candidates can gain Slack approve/reject actions before any email sends.
Multi-Role / Multi-Team Routing: One intake flow can fan out by department, each with its own requirements row, email copy, Calendly link, and Slack channel.
ATS / CRM Sync: Fields like fit score, status, matched required skills, and reasoning can push into HubSpot, Airtable, Notion, or a full ATS via API.
Hiring Analytics: Pipeline logs can feed Looker Studio or Power BI for pass-through rates, skill-gap trends, and time-to-screen reporting over time.