Mohammed Elsiddig

<p>Set the AI technical direction and stood up the function from zero.<br>
Selected and committed early to OpenClaw, forked and vendored it and the WhatsApp Cloud API plugin to lock builds.<br>
Owned governance, scoping, evals, cost ceilings, and security posture.<br>
Owned hiring for the function, including screening, interviewing, and onboarding AI engineers for Royal Furniture Group AI projects.<br>
Shipped and now operate the group's first production enterprise agentic AI operations system.<br>
Supported 3,500+ chat turns/day, 350-400 unique customer and staff users/day, and 1,400+ staff members on the system.<br>
Built secure role-verified control with proper scoping and hierarchy.<br>
Enabled staff and customers to interact through WhatsApp and a custom web application.<br>
Built the custom web application with React frontend and Node/Express plus PostgreSQL backend.<br>
Ran operations end-to-end including deliveries, order tracking, production status, scheduling/rescheduling, status updates, refunds/returns, customer-issue routing, driver delivery/route notifications, bottleneck detection, and department/team delegation.<br>
Implemented RLS-verified per customer and OTP-gated access for unrecognized phones using Twilio Verify.<br>
Created, updated, routed, resolved, and followed up on stale work.<br>
Assigned responsibility, absolved people when evidence showed a task was outside their scope, and recorded reasoning behind delegation decisions.<br>
Supported multimodality for files and media formats including images, documents, voice notes, locations, and contact cards.<br>
Tracked files and media across team and customer interactions and archived them for contextual recall.<br>
Produced bi-weekly and monthly reports for the CEO detailing staff performance metrics, ticket satisfaction ratings, AI-driven sales lead generation and lead conversion, and per-department KPIs.<br>
Evolved the system from an ad hoc OpenAI-based prototype into a custom agentic-loop architecture.<br>
Implemented API/provider fallbacks, budget control, recovery and error triage, agentic containment, and self-testing loops.<br>
Built a behaviour battery of 450+ scenarios across 19 dimensions to gate changes.<br>
Integrated the Hermes agent framework into the customer-facing AI system.<br>
Led migration from OpenClaw to Hermes.<br>
Used Hermes native self-improvement features including skill curation, memory consolidation, and background review.<br>
Used SQLite session store for session tracking and chat querying.<br>
Grew a corporate population of Hermes agents under an evolutionary algorithm for agentic population growth and control.<br>
Cut the model bill from a $900-1,200/month API-only OpenAI spend to about $200/month.<br>
Used an open-source Codex plugin customized for the agent framework running on a $200 OpenAI Codex seat.<br>
Used about $200/month AWS and other services and a $200 Claude Code subscription for auxiliary technical support.<br>
Implemented KV-cache hit rate tracking, frozen system/tool prefix, prompt caching, compaction routing, rf-llm-logger, CloudWatch spend dashboards, and daily alarms.<br>
Persisted every message, ticket lifecycle, heartbeat action, and per-LLM-call telemetry for review.<br>
Built the data pipeline by integrating the on-prem Orion ERP via AWS DMS into the agent's RLS-scoped Postgres mirror.<br>
Architected secure AWS end-to-end on least privilege using EC2, RDS PostgreSQL 16 with KMS and enforced SSL, CloudFront plus WAF, GitHub OIDC, SSM secrets, and Squid egress allowlisting.<br>
Owned CI/CD using GitHub Actions, ECR scan-on-push, and gated staging-to-prod.<br>
Built an internal finance/logistics research agent.<br>
Built a UAE FTA VAT-compliance agent for finance.<br>
Built a handwritten cost/BOM document-to-Excel pipeline with a human-in-the-loop review dashboard.<br>
Led the AI side of a robotic furniture-assembly program for the 6 factories.<br>
Worked on VLA action heads, JEPA-style world models, and NVIDIA Cosmos for sim-to-real.<br>
Shipped a Linux/Docker AI bench with YOLO11m-pose, MediaPipe, MQTT to InfluxDB to Grafana, and a FastAPI orchestrator.</p>

<p>Owned end-to-end delivery of Growbal/ActiveMile AI platform services across architecture, implementation, deployment, monitoring, and iterative optimization.<br>
Designed and shipped a cost-efficient agentic AI crawler for B2B company profiling.<br>
Achieved cost of less than $0.10 per company profile while sustaining 95%+ structured coverage.<br>
Built backend infrastructure with Django and PostgreSQL.<br>
Built AI logic with LangGraph, LangChain, DSPy, LlamaIndex, SerperAPI, and Claude.<br>
Implemented local fine-tuning and serving for GPT-OSS 120B using Hugging Face PEFT and vLLM.<br>
Deployed production-grade model hosting on Hyperstack.<br>
Built and productionized RAG and search pipelines including preprocessing, embedding generation, vector indexing, and low-latency retrieval.<br>
Implemented secure AWS infrastructure using EC2, Lambda, and RDS PostgreSQL.<br>
Enforced AWS best practices across IAM, credential handling, least privilege access, and environment isolation.<br>
Established observability-first operations with CloudWatch logs, metrics, and alarms.<br>
Integrated application and agent telemetry including DSPy execution logs.<br>
Implemented CI/CD across all microservices using GitHub and AWS deployment workflows.<br>
Drove throughput and latency improvements through workflow pruning, caching strategies, and modular service design.</p>

<p>Performed data collection, data version control, and data preprocessing.<br>
Built, trained, and deployed 2 deep learning models with accuracy above 98%.<br>
Incorporated pre-trained AI libraries and models such as AlphaPose, MediaPipe, and Detectron2.<br>
Built a custom Windows desktop application for AI services.<br>
Established and handled communications between the application and centralized governmental entities.<br>
Spearheaded end-to-end development of custom agentic AI applications for the Ministry of Defense.<br>
Conducted testing, debugging, and ongoing maintenance and improvement.<br>
Established rigorous logging and testing practices.<br>
Performed R&D on highly confidential production-stage computer vision applications.<br>
Deployed computer vision applications in 3 military sites.<br>
Solely managed end-to-end development of a suite of deep learning models from data collection and architecture design to training, production deployment, and enhancement.</p>

<p>Built a 3-fold Harmful Algal Bloom detection system.<br>
Used Google Earth Engine and Selenium WebDriver to generate satellite training data for one part of the AI detection system.<br>
Outsourced underwater microscopic image data and drone-captured image data to train the other 2 AI models.<br>
Performed data augmentation for outsourced data.<br>
Used OpenCV and Pillow to preprocess training data.<br>
Architected and trained 3 CNN models using TensorFlow and Keras.<br>
Incorporated DVC for data version control and MLflow for evaluation experiments.<br>
Issued initial PoC reports and usage documentation for Arias management and IoT engineers.<br>
Implemented feedback for system improvement and adjustment.</p>

<p>The construction of a 3 fold Harmful Algal Bloom (HAB) detection system.<br>
Using Google Earth Engine along with Web-Scraping tools like Selenium web driver to generate satellite training data for a third of the AI detection System.<br>
Outsourcing Underwater Microscopic image data along with Drone captured Image Data to train the other 2 AI models.<br>
Data augmentation for the outsourced data.<br>
Using Python computer vision libraries such as OpenCV and Pillow to preprocess the training data.<br>
Architecting and training 3 Convolutional Neural Network (CNN) models using Tensorflow and Keras. Incorporating DVC for data version control and MLflow for conducting evaluation experiments.<br>
Issuing initial PoC reports along with usage documentations for Arias management and IoT engineers as well as implementing feedback on the system improvement and adjustment.</p>

<p>Built payment verification backend structure for the company's personal finance manager app.<br>
Created and maintained enrichment API endpoints using Django REST API framework.<br>
Defined the structure and methodology for categorizing consumer transactions.<br>
Created KPI reports on enrichment accuracy using SQL and Python data manipulation and visualization tools.<br>
Used OpenCV to create a smart logo standardization mechanism for the enrichment service.<br>
Used Scrapy and Selenium WebDriver to populate the brand profiles database.</p>

<p>Built and visualized a datasphere that provides descriptive and predictive analysis on climate change data for a Jurisdiction Policy Network.<br>
Developed an AI tool/algorithm using electrical vehicle charging data from smart meters to characterize charging patterns and potential impact on the power grid network.<br>
Implemented exploratory data analysis on EV charging data using Python.<br>
Researched suitable machine learning models for temporal graph data.<br>
Implemented GNN-LSTM model training on graph data.</p>