Alper Can Özer
Community@Jessitoii · Bursa
I build AI agents and mobile ML apps. Sometimes I reimplement things from scratch to understand them.
Agent Skills by Alper Can Özer
Showing 9 vetted skills indexed across 1 GitHub repositories.
Jessitoiiscoring-engine
Scores job listings against a candidate profile using embedding filters and LLM analysis.
Jessitoiihitl-telegram
Implements Telegram-based human approval flows that pause and resume Celery job application tasks.
Jessitoiisession-lifecycle
Orchestrates daily job application sessions with state machine transitions, Celery tasks, and rate limiting.
Jessitoiidb-schema
Guides writing Alembic migrations, PostgreSQL queries, and ORM models for the CareerOS database schema.
Jessitoiiinterview-prep
Detects interview invitations and generates company research and prep documents.
Jessitoiiduplicate-prevention
Detects duplicate job applications using database hash checks, email parsing, and vector similarity.
Jessitoiicv-tailoring
Selects and reorders existing CV blocks to match job descriptions without fabricating content.
Jessitoiiats-adapter
Implements Playwright browser automation adapters for Greenhouse, Lever, Workday, and unknown ATS portals.
Jessitoiirate-limit-fallback
Implements LLM provider fallback chains with per-minute and per-day rate limit handling.
Frequently Asked Questions About Alper Can Özer
FAQPage SchemaWhat tasks can I automate using Alper Can Özer's skills?▼
You can automate the full job-application lifecycle: embedding and LLM-based relevance scoring, CV tailoring and cover-letter generation, stealth browser form submission on Greenhouse/Lever/Workday, duplicate-application detection, Telegram-based human approval, and interview preparation with company research.
Who are these skills designed for?▼
They target developers building or maintaining an autonomous job-application agent. Engineers working on LLM scoring pipelines, Playwright anti-bot evasion, Celery task orchestration, PostgreSQL/PGVector schemas, or Telegram human-in-the-loop flows will find directly applicable guidance.
What does the runtime architecture look like?▼
The system runs Celery-orchestrated daily sessions with a state machine (IDLE to DISCOVERY), rate limiting between platforms, and shutdown reporting. It uses PostgreSQL with PGVector and Alembic migrations, Playwright for browser automation, and a Groq/Cerebras/Ollama LLM fallback chain.
How does the system handle LLM rate limits and failures?▼
A dedicated fallback chain distinguishes per-minute from per-day rate limits, routing calls between Groq, Cerebras, and Ollama providers. It raises AllModelsExhausted when all providers fail and handles invalid JSON responses from LLMs during scoring.
What are the key prerequisites and dependencies?▼
Core dependencies include Playwright with stealth fingerprinting (Canvas/WebGL randomization, keystroke latency), PostgreSQL with PGVector and Alembic, Celery for task queues, Telegram bot webhooks for approvals, Gmail API for duplicate detection, and Groq/Cerebras/Ollama LLM access.