Ismail G.Private profile: Upwork shows its earnings only to signed-in members, so we check it every few weeks, not daily. MRR here is a rough estimate (Upwork's six-month earnings ÷ months active). Last read Nov 5, 2026.

I build forecasting and pricing models that tell you what to stock and what to charge, and I fine-tune open-source LLMs so you stop paying per API call. My latest forecast cut error by [ADD: X%] vs last year's sales. WHO I HELP - Businesses that need to decide what to stock and what to charge: e-commerce and Amazon brands, retail chains, B2B and SaaS companies, and service businesses like self-storage. - Teams paying for AI APIs in legal, insurance, finance and other fields, who want a private model that does their task as well as the API. FORECASTING AND PRICING - Demand and inventory forecasting: SKU-level forecasts, reorder points and stockout alerts, tested on your past months before you trust them. - Pricing analytics: price elasticity, what-if margin simulations and price test plans, so a price change is a measured bet, not a guess. - Predictive models: churn, propensity and lead scoring with XGBoost, with the drivers behind each prediction explained in plain language. LLM FINE-TUNING - LoRA and QLoRA fine-tuning of open-weight models (Qwen, Llama, Mistral) on your own examples. - Evaluation against your current API on a held-out set: accuracy, cost per 1,000 requests and speed, so you know if switching is worth it. - A forgetting check, so the model learns your task without losing what it already knew. - Deployment scripts to run the model on your own server or cloud GPU. PROOF - Inventory Copilot (portfolio project, public data): SKU-level demand forecast and reorder list on [ADD: dataset]. Forecast error [ADD: X%] lower than last year's sales in the backtest, with [ADD: N] stockouts flagged before they happened. - Private AI vs Paid API (portfolio project, public data): QLoRA fine-tune of Qwen2.5 on Banking77 (77 banking intents). [ADD: X%] accuracy vs [ADD: Y%] for the paid API, at [ADD: $X] vs [ADD: $Y] per 1,000 requests, and it kept [ADD: X%] of its general skills in the forgetting test. - Price Lab (portfolio project, public data): price elasticity per product with a what-if margin simulator. [ADD: main result, e.g. a price change worth +X% margin in the test data]. - Churn prediction (portfolio project): cleaned the public 7,043-row IBM Telco dataset, caught missing values hidden as blank strings in a numeric column, trained an XGBoost model ([ADD: ROC-AUC]) and ranked the top churn drivers. - 5-star Upwork review: "His work was accurate, well documented and always on time." - Store owner's view: I ran my own e-commerce brand, Spider Strike (millions of views, hundreds of sales), so I know which numbers an owner actually acts on. HOW I WORK 1. You send a sample of your data and the decision you need to make. 2. I reply with a plan, the baseline my work has to beat, and a fixed price. For a forecast, the baseline is last year's sales. For a fine-tune, it's your current API. 3. You get a first working version within 5 business days for standard projects, with a test report. 4. Handover: clean code, a short guide, and a sheet, app or script your team can use without me. WHY ME I show you where a model fails, not only where it works. If a simple method or your current API wins, I tell you, and you keep the budget. BACKGROUND Engineering student at Harbin Institute of Technology (Shenzhen), focused on AI and robotics. Python, pandas, SQL, scikit-learn, XGBoost, PyTorch, Hugging Face, Git, Linux. LANGUAGES AND HOURS English (fluent), French and Arabic (native), Chinese (conversational). Based in Shenzhen (UTC+8), online during US mornings and European afternoons. Send me a sample of your data and the decision you're trying to make. I'll reply with how I'd approach it and a fixed quote.