An AI pipeline that ranks opportunities and prepares applications
A Python workflow that gathers remote data jobs, removes duplicates, evaluates fit with several AI providers, prepares tailored documents and sends the strongest matches to Telegram.
Searching for relevant jobs means visiting several sources, reviewing repeated listings and preparing the same documents again and again. I wanted one workflow that could handle those steps and keep a clear record of every result.
My approach
How I approached it
The workflow applies simple rules before calling an AI model, stores every stage in Supabase and tries another provider when the first one is unavailable.
System flow
How the pieces connect
01Remote job sources
02Async fetch + normalize
03Supabase deduplication
04Deterministic prefilter
05LLM score + tailor
06DOCX + Telegram delivery
What I built
What is included
Async adapters for multiple job feeds with in-batch and database deduplication.
Heuristic prefiltering before LLM scoring to control cost and noise.
A Gemini, Groq and OpenRouter fallback sequence with usage-limit tracking.
Token, latency, success, and error telemetry for every model attempt.
Tailored résumés and cover letters rendered to DOCX with atomic writes.
Telegram delivery and persistent notification state to prevent repeats.
Key decisions
Choices that shaped the project
01
Use rules before AI
Simple rules remove obvious mismatches before any model call. This saves time and makes the first stage easy to explain.
02
Record every fallback
Provider changes, usage limits, response time and token use are recorded instead of being hidden inside a retry loop.
03
Guard against invented résumé claims
Returned highlights are checked against the structured profile; unsupported output falls back to a deterministic selection.
Result
What I delivered
The project shows a complete AI automation workflow with source collection, deduplication, model monitoring, document generation and delivery.
Next improvement
What I would do next
I would add a fixed evaluation set for scoring quality, reactivate the schedule and build a small monitoring page for source results, provider reliability, cost and application outcomes.
An interactive Power BI report that helps operations teams review battery life, compare consumption with historical averages, inspect individual devices and plan maintenance by location.
