Most of what I build doesn't have an interface anyone applauds — it has to log in somewhere, pull data on a schedule, and not fall over. Below is the actual body of work, grouped by what it does rather than what it's written in.
01
Python services that log into other people's interfaces — terminal portals, delivery apps, restaurant listings — on a schedule, and turn what they find into structured data.
The coordination layer for a container-appointment automation platform. Polls six freight-terminal portals — LBCT, Fenix Marine, Ports America, eModal/TermPoint, WBCT, Capyti — on a schedule, normalizes what comes back, and writes it to per-user storage. Long-running queries run in background threads and report status through a polling endpoint instead of blocking the request.
The bots underneath the orchestration layer — one automation script per terminal portal, each handling its own login flow and data extraction, exposed behind a shared API. Went through three rewrites as terminals changed their portals under us: a five-terminal FastAPI prototype first, then a larger Flask rebuild once the terminal list grew.
Captures full-page screenshots of terminal dashboards on request, then converts them to PDF and Excel through a LibreOffice conversion service — an audit trail for terminal data that has no export function of its own.
Shape shared by most of the automation work above — one bot per source system, one orchestration layer deciding when to run them.
A scheduler-driven API that scrapes food-delivery platforms — Talabat and Deliveroo live, Careem, Noon and Keeta scaffolded — every 30 minutes and caches deals behind one authenticated endpoint, with a small mobile client on top so checking five apps for deals becomes checking one.
A scraper that builds a structured dataset of restaurants and locations for downstream analysis, with a lightweight web frontend for browsing the results.
02
Systems where the interesting part isn't the API — it's what happens to the data once it's in: classification, synthesis, agents that act on it.
A decision-support system for a food-distribution company. It continuously ingests external signals — central bank data, food price indices, freight indices, news, financials — classifies and normalizes them into a live company model, and reasons over that model through a persona-calibrated decision agent. Built to a 14-day MVP timeline on off-the-shelf tools: sources were identified and verified as scrapable before any agent logic was written.
Reads WhatsApp sales conversations, works out which contacts, deals, tasks and notes they represent, and writes the result into HubSpot — so reps stop logging activity by hand. Prototyped as an n8n workflow, then rebuilt as a standalone Flask service with session management and sync analytics.
Takes a Figma file and produces a working, content-managed website — parses the design, generates templates, and wires them into a Grav CMS instance, with an LLM in the loop for content and code generation.
An MNIST classifier written from scratch — no PyTorch, no TensorFlow — to test a noise-exclusion training method: dynamic batching combining probabilistic sample selection with hard-negative mining. It reached ~41% recall on corrupted patterns at a low false-positive rate and converged 32% faster than a standard baseline while generalizing better.
03
Turning a dataset someone already has into something they'll actually look at.
A Power BI–style demo dashboard for an FMCG B2B client: dynamic KPIs computed from an embedded dataset, paired with a downloadable Excel workbook carrying the same measures as DAX formulas, so the client's own BI team could pick it up directly.
A client/product management tool with CRUD for both, plus exportable client, product, and summary reports through a small REST API.
github.com/mohamed-ali0/rahash →04
I studied computer & systems engineering at Assiut University. I work across a few clients at a time, mostly on automation, AI agents, and backend work that needs to ship fast and then keep running without anyone watching it.
Stack: Python, Flask, FastAPI, SQLAlchemy, Docker · JavaScript/React when a frontend's needed · C/C++, CUDA · n8n, LangGraph, CrewAI, MCP for agent work.
05