Ilyas Salas — Automation & AI Specialist · Rabat, GMT+1
Over 2 years in Amazon customer support for the French market. Now I automate the work I used to do by hand.
Designed and built by me. Each project details the problem, the solution and how it is built.
“My team answers ‘where's my order?’ a hundred times a day.”
An agent that checks orders and your policies, answers, and hands off to a human the moment it shouldn't act alone.
n8n · OpenAI · RAG (embeddings + vector store) · Google Sheets
“Prospective students compare driving schools in the evening, on Sundays, or when our line is busy.”
Conduite Atlas answers prospective students from the school's own information, books their appointments in Google Calendar and, when it can't answer, logs the request for a human to call back.
n8n · OpenAI · RAG (pgvector) · Google Calendar
“My sales reps keep calling numbers that don't exist.”
Every lead is checked the moment it arrives — phone, email, duplicate, missing fields — before it reaches the CRM. Every rejection is logged with its reason.
Make.com · Numverify · MillionVerifier · HubSpot
“Hot leads wait in the same queue as vague requests.”
Every request is scored 0–100 by AI, routed hot, warm or cold, created in HubSpot — and the prospect gets a personalized reply.
Make.com · OpenAI (structured JSON) · HubSpot · Tally
Over two years inside high-volume customer support taught me exactly where a process breaks at scale — now I automate it.
Systematic error handling and monitoring — alerts, automatic recovery and error logs on every scenario, beyond the usual no-code standard.
They answer repetitive questions from your own data, day and night, and pass anything that needs a human to your team.
Every request scored, summarized and routed, so your reps call the right prospects first.
Fake numbers, invalid emails and duplicates blocked before the CRM — with the reason for every rejection.
Contacts created and enriched automatically, alerts and follow-ups without copy-paste.
A CSV export is enough, even with fewer than 500 rows. Strip the names if you prefer: phone, email, campaign and date are all I need. You get your overall and per-campaign junk rate, with the reason for every rejection. The file is used only for this audit — numbers and emails only pass through the verification services — and I delete it once the report is sent.
