Systems Built
GrowEasy (Flagship)
AI-Powered ERP for SMEs. Modern full-stack SaaS platform for retailers, wholesalers, and freelancers. Integrated inventory, invoicing, and financial reporting designed to replace fragmented tools. Production-grade stack — now also the live SCM data layer behind Ahmad-AI's agent-queryable inventory insights.
Demand Forecasting Engine
Multi-model demand forecasting for wholesale distribution. Built on 84,741 real ERP order lines, 872 customers, and 4,307 SKUs. Three models (Naive, Seasonal Naive, ETS) selected per SKU via expanding-window backtesting. Reliability tiers, sensitivity checks, forecast-vs-trend sanity warnings, and an audit-ready Excel workbook with live formulas. Validated honestly against a 12-month holdout: the ETS model beat the naive baseline on 58% of pairs but lost 3× on volume-weighted WAPE because it extrapolated lumpy B2B order spikes as trend. v2 adds a prioritized replenishment worklist — a hard-coded trust rule refuses to auto-dispatch on any flagged SKU, and the worklist shows exactly what to order once the customer confirms.
Serverless Enterprise Warehouse Management System
End-to-end fulfillment system built entirely in Google Sheets and Apps Script. Handles the full cycle from receiving to delivery — auto-generating invoices, packing slips, shipping labels, and fulfillment records the moment a tracking number is entered.
Learn With Misham
Customer and Education Management Platform. Cloud-based LMS managing the complete student journey — online admissions across Academics, Quran, Naat, and IT courses, multi-student profiles, and an Anki-style progress tracker that pinpoints the exact words a student struggles with over time. Automated weekly reports keep teachers and parents in sync.
Ahmad-AI
Multilingual Quran & Hadith research agent. A retrieval-augmented Islamic knowledge system serving 113,842 indexed records — 6,236 Quran verses across 114 surahs, 50,762 hadiths from the Sunnah.com corpus, and 56,844 hadiths newly extracted from nine canonical Shamela collections (Bukhari, Muslim, Abu Dawud, Tirmidhi, Nasa'i, Ibn Majah, Muwatta, Musnad Ahmad, Darimi). Two-stage retrieval: paraphrase-multilingual-MiniLM-L12-v2 for recall across three ChromaDB collections, then GATE-Reranker-V1 — an Arabic-native cross-encoder — for precision reranking, with exact-citation lookup for surah:ayah queries. Built on a Shamela .bok → SQLite → JSONL → ChromaDB pipeline with SHA-256 deduplication, deterministic internal↔master book ID mapping, and per-book schema reconciliation across nine heterogeneous source editions.
Ahmad Voice AI
Real-Time Voice Agent. A continuous, multi-turn voice assistant that listens through the microphone, streams raw PCM over WebSockets, and speaks back in natural voice — no push-to-talk, no reconnect loops. Built on Deepgram's Voice Agent API, which orchestrates speech recognition (Nova-3), reasoning (GPT-4o-mini), and neural speech synthesis (Aura-2) inside a single managed session. Includes hardware-safe asyncio audio pipelines and a reconnect supervisor that survives transport drops mid-conversation.