Machine Learning Engineer at Evren AI, Level 2 Fiverr seller, and final year CS student. I design voice and chat AI agents, multi-agent workflows, and high-performing Meta and Google ad systems for brands around the world.
Final-year BS Computer Science student and Machine Learning Engineer at Evren AI, where I build production RAG agents, controlled-response conversational AI, and automation pipelines for real client-facing products.
My work sits at the intersection of two disciplines most people treat as separate. On one side, I engineer intelligent agents and automation systems with LangChain, LangGraph, n8n, Voiceflow and Botpress. On the other, I run Meta & Google ad accounts for Shopify brands, with a focus on getting the pixel, Conversions API and tracking actually right.
On Fiverr I am a Level 2 seller with a 4.9★ rating, trusted by repeat clients across the US, Canada, Europe, Australia and South Africa. Recent results: a 5.52x ROAS on a bottom-of-funnel retargeting build, and CPM cut from $87 to $32.50 through structured creative testing.
Pixel and CAPI setup, retargeting structure, creative testing, and ongoing optimization for Shopify brands.
PMax campaigns, conversion tracking cleanup, and policy compliance fixes.
Botpress and Voiceflow agents, plus RAG-based support and booking agents.
Connecting ad accounts, CRMs and APIs into automated pipelines that run without you.
Retrieval-augmented agents over your own docs and data, built with LangChain and LangGraph.
Fixing Event Match Quality, broken pixels, and Conversions API gaps that quietly cost ad accounts money.
Split bottom-of-funnel retargeting into three precise audiences and let the data show which one deserved the budget.
A 7-day creative test isolated two winning products. CPM went from an expensive guess to a proven number.
A traffic campaign aimed straight at the booking page for a Phoenix Airbnb host. Renewed for month two.
A RAG agent trained on the PAM 2006 contract, handling clause interpretation behind a gated client login.
Custom LoRA on RunPod for a WAN 2.2 image-to-video model, with ComfyUI workflow and lip-sync evaluation.
An n8n workflow that catches pixel and budget issues across live accounts before they cost money.
I highly recommend this fiver. mirzaumerikram understood what needed to be done better than I did. His work order was far more in depth than what I anticipated.
It was a real pleasure working with Meta Omer. He has in depth product knowledge, is solution focused, and works on a very timely basis. Highly recommended.
Excellent work, thoughtful, concise, detailed. Will work with again.
Worked on 3 different projects with him. Every single time, expectations were met and exceeded.
Excellent experience from start to finish. Highly professional, communicated clearly, and demonstrated impressive in-depth knowledge of the system.
Umer is a professional freelancer, he knows what he's doing and have good understanding of meta and how things work. I am continuing with him for a follow up.
Très bon travail de son côté, je recommande fortement.
Pleasure working with Meta Omer.
I've really enjoyed working with Mirza! Extremely efficient, knowledgeable and just a joy to work with. I'm looking forward to more work with him.
Mirza was very helpful and professional. He helped resolve the issues with my ad account, set up the campaigns, and adjusted the targeting based on my feedback. He was responsive and explained everything clearly. Thank you for your support.
Most underperforming ad accounts share one root cause: Event Match Quality so low the algorithm is optimizing blind. Fix the data before touching the creative.
From real client audits →If the knowledge changes weekly, retrieval wins. Fine-tuning locks yesterday's answers into the weights. Every production agent I have shipped uses retrieval first.
From production builds →Full autopilot fails silently. Every workflow I build drafts the action and waits for approval before anything touches money, customers, or live campaigns.
From n8n pipelines →