I turn complex business problems into systems, data and intelligent solutions.
I turn business data and processes into infrastructure for better decisions.
From data analysis to warehouses, dashboards and forecasting
Freelance — multiple clients
Asiatech
Eltiam Distribution & Kharazmi Pharmaceutical
Samanehaye Modiriat
Namnik Co. (Kik Ashna)
Malard Health Network
The tools I work with day to day
Each one ships full code, sample data and an artifact you can open
An interactive map of Tehran's 22 districts and Iran's 31 provinces, built from 516,947 real Divar listings and the Central Bank's transaction prices. District 1 asks 4× District 17 per square metre, and the cheaper districts rose fastest. Enter a budget to light up the neighbourhoods that fit, price a home with the in-browser model, and move the 12-month forecast with your own growth assumption.
Customer segmentation, a churn model trained on a time cutoff and a credit-risk model (PD/LGD/EAD), alongside a monthly P&L, a 12-month forecast and marketing-budget optimisation: how the bank grows customers and loans while keeping credit risk in check.
A Dynamics 365 Sales export (Dataverse tables with option-set metadata) goes through 71 quality checks into a star schema: what each marketing channel returned, why deals are lost, margin per ton. A win model ranks deals better than the probabilities reps type into the CRM.
780K rows of sales and Dynamics 365-style CRM data. The quantile model treats the mega-sale calendar as a feature, not noise, and the churn model is trained on a time-based split rather than a random one.
ABC/XYZ segmentation decides which SKU deserves safety stock and which is just parking capital. Quantile forecasts set the service level against the tail of the distribution, not the mean.
One step ahead, XGBoost and SARIMA are close. Over a 45-day horizon XGBoost's error grows 14-fold, because trees cannot extrapolate a trend — a result that makes the evaluation method matter more than the model. One step ahead, SARIMA reaches an R² of 0.81.
Every open CRM opportunity enters the forecast weighted by its probability. Capacity across four lines is solved as a linear program constrained by machine and labour hours, and MRP explodes a real BOM into purchase orders.
Official unemployment fell from 12.4% to 7.5% while participation dropped 2.6 points: the statistic improved because more people left the labour market. A bilingual story with a model whose assumptions you can change.
Orders were growing 65% a year while contribution margin stayed under 7%. A per-city P&L, cohort analysis, a 12-month forecast and a constrained reallocation of the promotion budget show where growth turns profitable.
A forecast is not a production plan. Here the Holt-Winters output passes through the capacity of five lines and lands on a schedule that is actually feasible, with an MRP for raw materials.
Built on real, citable World Gold Council and IMF data. Twenty-six years of reserve history show the story is not the leaderboard — it is the buying flow after 2009.
A fully tied three-statement model whose balance sheet balances to the penny, a 12-month revenue, EBITDA and cash-flow forecast, and an explainable budget-overrun risk model — for the same telecom operator as the HR project.
Every employee gets an attrition-risk score with published weights, not a black box. The source of truth is a live-formula Excel workbook, and the output is a 12-month hiring plan per department.
A marketplace sales forecast becomes a day-by-day staffing plan per distribution centre: how many pickers, how many riders, on which day and at which hub.
EV production, trade and the battery supply chain, chapter by chapter on the IEA Global EV Outlook, with an interactive world map and a Python-generated Excel workbook.
Operations analysed through throughput, service level and capacity: which unit carries the load, where waiting time breaks, and which metric actually predicts a delay — with a dashboard that puts those numbers where the daily decision is made.
Raw user events land in the warehouse through incremental loads and data quality gates, alongside a partitioned Parquet table built for fast analysis.
A daily sales dashboard for an online store: revenue, profit, basket and customer behaviour on one page, with period-over-period comparison on a proper star schema. The PBIX file ships with it, so you can open the model, the relationships and the DAX measures — not just look at a screenshot.
A free Persian reference of data and BI interview questions — SQL, modelling, Power BI and statistics — with answers that explain the reasoning instead of asking you to memorise it.
Open the question bankBu-Ali Sina University, Hamedan: tutorials on inventory models, reorder points and safety stock — the same concepts I now apply in production planning projects.
Hands-on training in Power BI, data modelling and building a first data warehouse, for teams that want to own their reporting. In Persian or English.
Ask about a sessionDesigning a sustainable EV battery supply chain under uncertainty with a data-mining approach: demand and return scenarios are generated from data rather than from arbitrary assumptions.
A review of data-mining methods for breast cancer prediction — the first place I learned that model accuracy without a baseline is a meaningless number.
For a role, a project, or training for your team
