AI Recruiter
Intelligent job discovery, candidate evaluation, and automated recruiter outreach platform.
01. Problem & Context
Modern job searches require tracking dozens of companies, reading complex job descriptions, and personalizing outreach to verified hiring managers. Most tools either lack intelligent fit analysis or automate email spam without human review.
02. Engineering Solution
Built a candidate-first platform pairing a Python 3.12 FastAPI backend and Celery/Redis asynchronous queue with a React/TypeScript dashboard. Features provider-neutral LLM evaluation of candidate alignment, DNS MX-record validation for recruiter discovery, and an ironclad database-enforced human approval gate for all Gmail outreach.
03. System Architecture
The system decouples asynchronous scraping, LLM evaluation, and recruiter verification via Celery worker queues and Redis brokers. A FastAPI service exposes RESTful endpoints with Argon2/JWT authentication and AES-256-GCM encrypted credential vaults, consumed by a reactive Vite + React dashboard.
Core Architectural Components
Provider-Neutral AI Evaluator
Structured LLM evaluation assessing candidate-job alignment, required vs. preferred skill matching, and token/cost accounting.
Recruiter Verification Engine
Identifies technical recruiters and hiring decision-makers with evidence-based DNS MX-record checks and confidence scoring.
Human-in-the-Loop Outreach
Drafts personalized email sequences via authorized Google Gmail APIs with a mandatory database-enforced human approval gate before dispatch.
Asynchronous Worker Queue
Celery 5.4 and Redis 7.2 manage long-running background scraping, evaluations, and delivery tasks without blocking HTTP handlers.
Encrypted Credential Vault
AES-256-GCM authenticated encryption protects OAuth tokens and integration keys in PostgreSQL with zero plaintext leakage.
04. Notable Engineering Decisions
Ironclad Database-Enforced Human Approval Gate
Autonomous outbound email risks reputational harm and provider suspension. Communications can only transition to 'sent' through explicit, authenticated user approval in the dashboard.
Asynchronous Background Processing
LLM evaluations and DNS MX validations introduce variable network latency. Celery workers and Redis queues isolate background work from synchronous user requests.
Provider-Neutral AI Layer
Abstracting LLM calls behind structured evaluation contracts enables seamless switching between OpenAI, Anthropic, or local open-source models (via Ollama or llama.cpp) without changing domain logic.
05. Engineering Challenges & Solutions
Challenge: Preventing false positives and deliverability bounces in recruiter discovery.
Engineering Solution: Integrated DNS MX-record validation, corporate domain format pattern heuristics, and multi-factor confidence scoring before generating outreach targets.
Challenge: Securing third-party OAuth tokens and personal communications data.
Engineering Solution: Implemented application-level AES-256-GCM encryption for stored Gmail API credentials and Argon2 password hashing.
06. Implementation Highlights
- Python 3.12 FastAPI backend with asynchronous SQLAlchemy 2.0 and PostgreSQL 16
- Celery 5.4 worker queues backed by Redis 7.2 for reliable asynchronous tasks
- React 18 + Vite dashboard with TypeScript, Tailwind CSS, and TanStack Query
- AES-256-GCM authenticated encryption for credential governance
- 18-stage recruiting pipeline tracking opportunities from discovery to offer
- Full Docker Compose deployment with automated health checks