Mahmoud Yosristuff maker
All projects
case study, 2026

Argon

A job-search assistant that collects job ads from everywhere, scores each one against your CVs, and turns the whole market into a plan.

Role
Designed and built end to end
Status
Private beta
Runs on
Desktop and phone
Markets
Egypt and the US
argon / jobs
Argon Jobs page: ads grouped by week, each scored against the best CV
every ad scored against your best CV
grouped by week, newest first
the problem

In Egypt, the jobs are in the chats

A big share of tech job ads here never reach a job board. They get posted in Telegram channels and WhatsApp groups, scroll away in a day, follow no format, and often give no way to contact anyone. People read hundreds of messages, guess which ones fit, and miss the good ones because they were posted at 2 AM.

I wanted one place that reads all of it for you, tells you which jobs fit your CV, helps you apply fast, and shows you what to learn next based on real ads, not guesses.

How it works

from a messy feed to a plan
1

Collect every ad

Argon imports ads from Telegram channels, WhatsApp groups and job sites, then cleans them into one list. An admin-only job engine adds fresh listings so Argon doesn't depend on chats alone.

  • Every ad is matched to a field pack (AI, marketing, sales and more), so Argon works for any field.
  • Filters for location, work mode, contact type and minimum score.
the dashboard: jobs, applications, replies and a year of activity
the dashboard: jobs, applications, replies and a year of activity
2

Score each ad against each CV

Upload one or more CVs. Argon scores every ad from 0 to 100 against each one and shows the best CV for that job, so you know which version to send.

  • Scores recalculate after you edit a CV.
  • The same view works on a phone.
jobs in dark mode
jobs in dark mode
every application and where it stands
every application and where it stands
jobs on a phone
jobs on a phone
what to learn, on a phone
what to learn, on a phone
3

Apply, then keep track

Argon drafts applications, builds tailored CVs and fills common forms. Then it watches your mailbox.

  • It finds applications you already sent and suggests which job each belongs to. Nothing is linked until you confirm.
  • Replies from companies attach to the right application.
  • A GitHub-style graph shows a year of activity.
sent emails matched to jobs, and replies picked up
sent emails matched to jobs, and replies picked up
a year of applications
a year of applications
4

Read the market

The Market page analyses every ad from the last 90 days and writes a plan with the evidence behind each step.

  • From real data: 53% of AI ads were posted between 22:00 and 04:00 Cairo time, so importing right after puts you among the first applicants.
  • What to learn next ranks skills by demand, by how many jobs each opens for you, and by trend.
  • A note on how far to trust the numbers shows the gaps in the data.
six steps to get hired, each linked to evidence
six steps to get hired, each linked to evidence
what to learn next, ranked
what to learn next, ranked
skills the whole market asks for
skills the whole market asks for
5

Labs: companies that just raised

Companies that just raised money usually hire in the next months, often before they post. Labs tracks funding news, scores how well your skills fit each company, and helps you write a short note to the team.

funded companies
funded companies
how I can help: match and a note draft
how I can help: match and a note draft
employers that suit your CVs
employers that suit your CVs
6

Admin: learns, and stays safe

  • Skill map: each field pack weighs skills by what real ads ask for, and updates from new data.
  • Pack changes: Argon proposes new skills and weights. A person approves each one.
  • Backups: scheduled, encrypted with AES-256, restore into SQLite or PostgreSQL.
  • Local AI: an Ollama model can run on the same machine, so CVs never leave it.
skill map with learned weights
skill map with learned weights
pack changes waiting for approval
pack changes waiting for approval
encrypted backups
encrypted backups
resource use
resource use

Why it's built this way

  • Field packs, not hard-coded techSkills, weights and spellings live in data packs. A new field is a new pack, not new code.
  • Suggest, never act aloneEmail matches, pack changes and drafts wait for the user to confirm.
  • Show the evidenceEvery number in the plan links back to the ads behind it.
  • Private by defaultCVs and tokens are encrypted, and the AI can run locally.
  • Near-zero running costRuns on a free cloud tier or a home server.
  • Any screen, any lightLight and dark themes, desktop and phone layouts.

The code is private while the beta gets ready. Screenshots use demo data, so names like "Acme Vision AI" are placeholders.

next on the benchRubies