Stop updating spreadsheets manually. Connect your Gmail securely and let our intelligent agents automatically extract, track, and organize your job applications and interviews.
Built from the ground up to understand the nuances of modern recruitment emails.
We utilize Groq's blazing fast Llama-3 endpoints to accurately pull company names, roles, and context from unstructured emails.
Your data is yours. Authenticated directly through Google OAuth, isolated by tenant, with no passwords stored.
Our pre-filter constantly learns. Pure spam and newsletters are ignored aggressively, keeping your dashboard pristine.
Every email passes through a pipeline of specialized AI agents — each with a single, well-defined job.
Ingestion
Securely polls your Gmail inbox via OAuth. Fetches up to N recent emails, strips HTML, extracts headers. Concurrent processing via ThreadPoolExecutor — multiple emails are handled in parallel.
Gatekeeper
Pure keyword matching — no LLM call. Scans subject + body against a curated list of hardcoded and learned signal phrases. If a keyword matches → email proceeds to the classifier. If nothing matches → immediately ignored, saving 100% of the LLM cost for that email.
Classification
The LLM brain. Uses Groq's Llama-3.1-8B to read the email and output a structured classification: type, company, role, recruiter name, and a candidate keyword for learning. Guided by a dynamically built prompt that includes past mistakes from the guide file.
Specialist Agents — Orchestrated by LangGraph
Reads company and role from the classifier output. No extra LLM call — just sets status to applied and enriches metadata.
Extracts interview date/time, timezone, type (phone/technical/onsite), interviewer names, meeting link, and duration.
Pulls platform (HackerRank, CodeSignal…), assessment link, time limit, due date, and test type from the email.
Detects emails requiring action (EEO surveys, email confirmation). Extracts the verification URL, description, and deadline.
Extracts rejection stage (screening/technical/final round), stated reason, and whether future applications are encouraged.
Handles follow-up nudges. Extracts the action needed, deadline, and any meeting link. Maps to interviewing status.
Persistence
Pure database work — no LLM calls. Matches the email to an existing Application using a 3-tier strategy (Company + Role → Company + Sender → Company fallback), then creates or updates the record. Also indexes the email in ChromaDB for future semantic search.
Self-Learning — Background Tasks
After a successful LLM classification, the classifier emits a candidate keyword phrase. This agent asks the LLM to score it from 0–10 — how universally job-signal is it?
Company names and job titles are scored low (1-3) and never saved. Universal phrases like "coding assessment" score 9-10.
Triggered whenever you manually correct a category. Analyzes the email body, understands what the LLM got wrong, and writes a permanent rule to the guide file.
The guide file grows over time. Every future LLM call includes these rules, making the system smarter with each correction.
Every sync is database-first. Your tracker reflects the exact state of your PostgreSQL database — any UI action commits to the DB before updating the interface. A full page refresh is always consistent.