Hi, I'm Ahmed Hashim
My flagship is My Stream — an AI publishing platform that drafts articles in your editorial voice using Claude, GPT, Gemini, and four more providers. I also build digital Mushaf and Quran apps used on web and Android. All shipped, all maintained, all here.
Model
Providers
Editorial voice
Voice
Journalistic
Reading level
General
Words
~ 900
Min quality
72
Today's run
Schedule
Weekdays · 09:00
Next run in 14h
Sources
About
I'm a software engineer who likes the whole pipeline — from a half-formed idea to a shipped, maintained product. My current focus is applied AI for publishing workflows, but I also keep a long-running line of work in Islamic resources: a digital Mushaf and a Quran Android app that have been in users' hands for years.
Most of what I build is shipped solo. I'm comfortable owning the stack end-to-end: Angular and Ionic on the front, AWS and Firebase on the back, with whichever AI provider fits the job. I write occasionally about what I'm learning, and I'm open to consulting on AI-first editorial tooling.
Tech I reach for
What I'm shipping
Three products I build and maintain — one flagship, and a long-running line of Islamic apps.
AI publishing platform
Scrape sources, score with quality thresholds, and draft articles in your editorial voice. Multi-model (Claude, GPT, Gemini, Grok, DeepSeek, Mistral, Groq), workspaces, scheduling, BYOK.
A clean digital Quran reader
A focused, distraction-free Mushaf for the web at mushaf.geohashim.com. Built for fast page navigation, comfortable reading typography, and a layout that respects the printed Mushaf.
A Quran reader for Android
My long-running Quran app for Android, available on Google Play. Reciter audio, bookmarking, and offline reading — for everyday use.
Architecture
Every published article walks the same five stages. You configure each stage once per workspace; the schedule does the rest.
Stage 1
RSS · topics · feeds
You configure
Feeds + topics
Stage 2
Lambda
Stage 3
Model picks one
You pick
Model · threshold
Stage 4
Voice + length
You define
Voice · style
Stage 5
Stream + share
You ship
Review + publish
Sources. You point My Stream at RSS feeds, topic strings, and (soon) custom HTML scrapers. Each workspace has its own list.
Scrape. A Lambda fans out, fetches each source, normalises the HTML, and queues candidate items.
Score. The picked model rates every item against your editorial rubric. The threshold is per-workspace; the model offsets auto-correct so cheaper models do not silently fail.
Draft. Survivors of the score gate are drafted in the voice you defined: voice, reading level, target word count, target keywords, citation rules, custom prompt additions.
Publish. You review the queue, edit if you want, then publish to your stream. Every post gets a permalink, an OG card, and a share-button row.
Hover any stage to see what it does. The dots flowing between stages are real units of work — sources, scored items, drafts.
Editorial voice
The drafter respects an editorial voice you define per workspace. Click a voice to see the same source rewritten — this is exactly the prompt shape My Stream uses.
AI safety moved from back-office concern to front-page issue this year, with a wave of frontier-lab disclosures forcing every team that ships an LLM-backed product to revisit its deployment guardrails. Independent analysts say the shift is less about model capability and more about the pace at which safety teams are being asked to certify production releases.
Same source. Same topic. The drafter swaps voice, reading level, and length per workspace.
Multi-model
Anthropic runs out of the box. The other six plug in via bring-your-own-key — costs land on your provider bill, the prompt stays the same.
3 models
Best default — runs without keys.
4 models
Reasoning-tier picks excel at analysis.
3 models
Cheapest tier with a usable context window.
2 models
Strong on current-events grounding.
2 models
Reasoning at a fraction of the cost.
2 models
EU-hosted option for compliance.
1 model
Wildest token throughput on the market.
BYOK = Bring Your Own Key. Drop a provider key into a workspace and the same prompt routes through that provider — costs land on your bill, not ours.
Live demo · runs in your browser
Pick a model, set a quality threshold, type a topic, and hit Run. Watch the same four-stage flow My Stream uses for real — scraped, scored, drafted, published — execute live.
Scrape
pending
Score
pending
Draft
pending
Publish
pending
Hit Run pipeline to see a draft preview here.
This is an animated mock — no real LLM call leaves your browser. The numbers, model offsets, and stage flow mirror what My Stream actually does.
An end-to-end editorial pipeline — scrape, analyze, score, draft, publish — wired to whichever frontier model fits your budget.
Pick from Anthropic, OpenAI, Google, xAI, DeepSeek, Mistral, or Groq per workspace. Anthropic runs out-of-the-box; the rest plug in via BYOK.
Drop your provider keys into the BYOK panel. Costs land on your bill, not ours, and the same prompt routes through whichever model you pick.
Voice, reading level, target word count, target keywords, citation rules, custom prompt additions — set per workspace and the drafter respects them.
Every draft is scored against your minimum quality threshold. Cheaper models score lower on the same rubric, so the threshold auto-adjusts per model — no manual retuning.
Run the pipeline daily, on weekdays, or never. The automation Lambda streams progress over GraphQL subscriptions so you watch a job execute live.
Each org has its own model, voice, source feeds, keywords, and schedule. Switch workspaces without losing context.
“I get a daily queue of properly-sourced drafts in our voice. The bit that used to take a writer four hours now takes me twenty minutes of editing.”
My Stream for AI publishing. Mushaf for the web. Online Quran for Android. All linked above.
Open My Stream