Ahmad Saad

Architected and solely built a production trend intelligence platform in TypeScript, Node.js, and PostgreSQL, running on the official TikTok API for Business alongside commercial data vendors. It measures rising sounds, formats, and topics across 8 discovery lanes and publishes an evidence-grounded dashboard for the regional marketing team.

Measurement decides, LLMs describe: deterministic signals such as caption n-gram clustering, creator spread, adoption curves, and freshness decide every verdict. A model can name a mechanic or demote a measured riser, but it can never admit an entity the measurements rejected, and it never writes URLs, IDs, counts, or provenance.

Two-stage screening: a Claude-based brand-fit gate applies voice and brand guardrails and returns structured go or no-go verdicts, followed by a computer vision and reasoning layer that scores GCC cultural applicability. Works in Arabic and English.

Cost governance: per-run spend meters plus a cross-process daily budget ledger in PostgreSQL, with balance-gated runs. Thresholds self-calibrate from each lane's trailing spend history rather than hand-tuned constants. Combined data and inference spend holds to a fixed single-digit-dollar daily budget.

Full stack ownership: a Dockerized collector running 5 scheduled cron lanes, a Next.js dashboard on Cloudflare Workers behind Cloudflare Access, and custom edge middleware for per-IP rate limiting and CSP headers. 315 automated tests, 16 versioned SQL migrations, npm workspaces monorepo.

Forward deployed working style: scoped directly with marketing, creative, and innovation stakeholders and iterated in short feedback cycles. Replaced external tooling with internal trend research, reducing an estimated $60K in spend compared to similar platforms, and supported regional launch campaigns, contributing to a projected 20% increase in engagement rates and recognition from regional HQ.

Supported SCADA and substation automation projects across Siemens medium-voltage switchgear platforms (NXAIR, NXPLUS C, 8DJEH). Covered EPC execution workflows, technical submittals, procurement, and on-site material inspections for large-scale water infrastructure.

Directing a company-wide GTM strategy for a client, scaling operations and restructuring revenue streams. Rebuilt client workflows across process management tools.

Restructured client workflows in ClickUp, reducing operational inefficiencies by 25%. Supported the CEO with decision-ready operating insights.

Shipping production-grade, executive-facing WordPress and Elementor websites to live domains in under 48 hours, translating client requirements into design systems, implementation, testing, audits, and deployment orchestrated end to end across AI agents.

Building sites that non-technical teams maintain themselves: structured content, self-updating Elementor listings, and editor documentation, so no one edits code or layout to publish.

Delivering mobile- and web-optimized, accessible (WCAG AA), performance-tuned front ends with purposeful animation following UX/UI best practices, plus automation that pulls in images and metadata, scoped against documented client requirements.

Developing and maintaining the CITL AI Hub, a dynamic WordPress site used by AUS faculty: led its HTML and CSS redesign against stakeholder requirements, prototyped in Figma, and handle ongoing updates for content accuracy, functionality, and engagement. Researching emerging AI technologies and producing written evaluations that shape faculty adoption.

Speech emotion recognition: classified emotion from audio snippets using log-mel spectrogram representations, fine-tuning ResNet and MobileNet backbones in TensorFlow and Scikit-Learn. Reached roughly 75% accuracy after augmentation and hyperparameter tuning.