Danish
Ahmed
I ship production software end to end, from database schema and security model to containerised deployment and a passing test suite: a live e-commerce platform, multi-agent LLM systems and WhatsApp agents for clients in the US, Latvia and Pakistan.
Shipped
What I Do Best
Full-stack engineering and applied AI, from database schema and secure checkout to LLM agents, RAG and production pipelines. I build the whole system and test it.
Professional Experience
From LLM evaluation at Turing to architecting production AI agents for European e-commerce, every role sharpened my ability to build systems that work at scale, in the real world.
Present
- Sole engineer on a production e-commerce platform: content model, storefront, checkout, admin panel, blog and a dual taxonomy of categories and curated collections.
- Security-first checkout (server-side price recomputation, hashed OTP, timing-safe webhook verification, rate limiting, RBAC); a security audit with 31 verified fixes and 9 false positives rejected.
- 56 Playwright end-to-end tests and 245 Vitest tests; eight environment-gated integrations and seven authored n8n order-lifecycle workflows. Payment and courier go-live pending merchant onboarding.
Present
- Built a WordPress / WooCommerce storefront for the full guide catalogue: AJAX filtering and live search, six retailer buy-links per guide, and a three-state crowdfunding hub the client switches from an admin bar.
- Engineered a retailer-link pipeline across 5+ storefronts: scraped 409 Google Play titles with a fuzzy-matching engine, matched 304 PublishDrive titles with an audit log, reconciled ~230 uncatalogued titles and audited 198 IngramSpark titles.
- Built an OCR cover-QA system over ~900 cover images (title-vs-filename mismatch, letterbox and aspect detection, best-source mapping).
- Architected and deployed OpenClaw, a 24/7 AI publishing assistant on a hardened Hetzner VPS (Ubuntu 24.04) in Docker, integrating Anthropic Claude API with a Telegram bot; delivered all 27 Phase 1 deliverables in 3 days vs. a 2-week estimate.
- Hardened the server end-to-end: Tailscale private networking, SSH key-only auth, UFW firewall, Fail2ban brute-force protection, non-root Docker user, zero credentials stored in config files.
- Built Phase 2, 13 production features: Gmail + IONOS email read/reply via Telegram, voice-to-draft (OpenAI Whisper), OneDrive integration, catalogue intelligence over the full guide catalogue, royalty monitoring across multiple platforms, weekly trend reports, and daily morning briefings.
Jun 2026
- Performed side-by-side evaluations of large language model outputs for quality, accuracy and relevance against established evaluation guidelines.
- Performed systematic hallucination detection and fact-checking, documenting structured justifications for every evaluation decision to build a traceable audit trail for model improvement cycles.
- Worked a four-hour daily overlap with US Pacific time, using browser-based tooling and command-line instructions, and documented findings for stakeholders.
Dec 2025
- Architected and shipped a production WhatsApp AI agent for a Latvian e-commerce client using n8n and Meta WhatsApp Cloud API, automating lead qualification, real-time inventory lookup via Zoho Sheets, and PDF offer generation, with human handoff for payment.
- Designed multi-stage conversational NLP workflows routing inquiries into Simple (direct offer) vs. Complex (custom spec + approximate invoice) tracks with automated human handoff triggers.
- Delivered a complete handover package: schema definitions, deployment runbook, integration guides, and UAT support, verified working across the full end-to-end flow.
Nov 2025
- Built a fully serverless end-to-end AQI forecasting ML system: automated real-time ingestion from OpenWeather API, engineered 29+ features (lag values, rolling stats, cyclical time encodings, pollutant ratios), stored in Hopsworks Feature Store.
- Trained and benchmarked 4 models (Random Forest, Gradient Boosting, Ridge, Lasso), Ridge Regression best: RMSE 9.14, MAE 7.53, R² 0.741; deployed hourly feature pipelines and daily automated retraining via GitHub Actions CI/CD.
- Shipped an interactive Streamlit dashboard with live 72-hour AQI forecasts, SHAP-based model interpretability, hazardous AQI alerts, and 7-day historical trend analysis.
Present
- Developed and deployed production client websites (wbbre.com and whiteweldfoundation.org in 2025, with interface design by a designer; storybookhomesnj.com in August 2026), fully responsive and live.
- Building AI-integrated WordPress e-commerce sites for family businesses transitioning from physical to online retail.
Aug 2024
- Documented 9 institutional Power BI dashboards (SCCAF, Teaching Allocation Model, Teaching & Learning, Strategic Plan Monitoring, MBBS Curriculum Mapping, QTL_net Surveys, Alumni Survey 2023, Student Evaluation of Teaching, Pipeline of Student Journey).
- Wrote user guides and technical documentation covering Power Query, DAX and data-model logic for each dashboard.
- Produced a narrated tutorial video for the Student Evaluation of Teaching dashboard and presented the documentation to the Director of IDAR (27 Aug 2024).
Jun 2023
- Rendered 1,127 volunteer hours (19 Sep 2022 - 26 Jun 2023, confirmed by an AKU HR letter).
- Researched and updated alumni records for AKU's worldwide alumni community (roughly 10,000 at the time) in the Raiser's Edge database and Excel, finding each alumnus's current position and university, matching them to LinkedIn profiles and recording the details and profile links.
- Entered applicants' details for alumni-card processing, handling confidential data with administrator access under staff supervision.
- Assisted staff at Alumni Affairs events including PGME graduation, reunions for SONAM, IED and the Medical College (MBBS), the University convocation, the Impact Symposium, internship ceremonies, the Distinguished Alumni Awards and the AKU 40-year programme.
Selected Projects
Blink, AI Policy Assistant
Final Year Project · RAG System with a Local LLM · Team of four
A RAG-powered policy assistant with a locally run LLM for enterprise environments. Team of four: I built the policy data, the SQL Server schema and the RAG pipeline (embeddings, FAISS retrieval, LLaMA Phi-3 via llama.cpp), and ran task planning and the report; teammates built the Windows client, dashboards and security. FastAPI backend with a .NET desktop frontend, role-based answers scoped to each user tier, and audit logging. After the FYP the team added a React / Express / MongoDB web client on the same local-LLM backend.
OpenClaw, AI Automation System
24/7 Autonomous Publishing & Business Assistant
Architected and deployed a production AI assistant for a US travel publisher, integrating Anthropic Claude API with Telegram, Gmail, IONOS, OneDrive, and OpenAI Whisper on a hardened Hetzner VPS. Delivered all 27 Phase 1 deliverables in 3 days vs. a 2-week estimate. Phase 2 added 13 live features: voice-to-draft, catalogue intelligence over the full guide catalogue, royalty monitoring across multiple platforms, and automated daily briefings.
in 3 Days
Features
Operation
Platforms
WhatsApp AI Agent — E-Commerce Client
Production E-Commerce Automation · Latvia
End-to-end WhatsApp AI agent for a Latvian e-commerce brand. Automates lead qualification, real-time inventory lookup via Zoho Sheets, PDF offer generation, and human handoff for payment; standard orders are handled automatically up to that point. Multi-stage NLP routing for Simple vs. Complex orders.
Healthcare Translation Web App
Real-Time Medical Interpretation Tool
Real-time speech-to-text medical translation app with dual-language transcript view, TTS playback, and mobile-first UI. Vanilla JavaScript on Vercel; Urdu, English, Hindi and Arabic, with a multi-API translation fallback and an optional OpenAI step. A working prototype, not a clinically validated device.
AQI Forecasting ML System
Serverless End-to-End ML Pipeline
Fully serverless AQI forecasting system: 29+ engineered features, Hopsworks Feature Store, hourly pipelines, and daily automated retraining via GitHub Actions CI/CD. Ridge Regression achieved RMSE 9.14, R² 0.741. Live Streamlit dashboard with SHAP interpretability and 72-hour forecasts.
Barakah Abaya, E-Commerce Platform
Production Next.js / Payload CMS Storefront
Solo-built on Next.js 16, React 19 and PostgreSQL: hardened checkout, eight integrations, a security audit with 31 verified fixes, and 56 Playwright plus 245 Vitest tests. Live at barakahabaya.com; payment and courier go-live pending merchant onboarding.
Agent Core, Multi-Agent LLM Framework
Personal Platform · Python, MCP v2
Provider router with retry and cascading fallback, checkpointed execution that resumes after a power cut, a multi-agent crew with BM25 memory, a bi-temporal knowledge graph, and deterministic honesty, PII-redaction and refusal gates. Includes a scheduled lead monitor across 37 sources. Source code is private.
WhatsApp Real-Estate Lead Agent
Lead Qualification · 2026
Deterministic state-machine agent that qualifies property leads over WhatsApp and runs keyless from the CLI, with Dubai (AED) and Karachi (PKR) locale modes. Public on GitHub.
About Me
I'm a full-stack and AI systems engineer from Karachi with a focus on building real, production-grade AI systems, not just prototypes. My work spans LLM orchestration, ML pipelines, data analytics, and backend integration.
What sets me apart is an engineer's pragmatism combined with a researcher's curiosity. I've shipped AI products for Latvian e-commerce, a US publishing house, and university departments, each with different constraints, each demanding a tailored solution.
I studied Computer Science at KIET (coursework completed Fall 2025) and work in English and Urdu, with German in progress. I'm open to international opportunities globally and welcome visa sponsorship.
Ready to Build
Together?
Whether you need an AI agent, a data pipeline, or a machine learning system built right, reach out. I respond within 24 hours.
Based in Karachi, Pakistan.
Available immediately · Visa sponsorship welcome.
Open to remote, hybrid, or relocation worldwide.