AI Engineer / Houston, TX

Portrait of Ahmad Ferdaws Shafiq with the Washington Monument behind him

profile / ferdaws

ai/ml engineer

Ahmad Ferdaws Shafiq

CS Undergrad / AI Engineer / Houston, TX

AI Engineer focused on systems that ship. I design and build production agentic pipelines, ML models, and automation infrastructure that solves real problems at scale.

View Projects Get In Touch Resume

About

Builder Profile

I'm Ahmad Ferdaws Shafiq, a Computer Science student at North American University and AI Engineer based in Houston, Texas. My interest in technology started early and never stopped growing. Over time that curiosity became a discipline: building systems that are technically rigorous, production-ready, and actually useful.

Today I work across agentic AI, machine learning, and automation infrastructure, and I'm pursuing graduate research to push further into what's possible with AI systems.

90%

Manual workload reduced

1K+

Documents per week

40+

Staff hours saved / cycle

Degree: B.S. Computer Science
School: North American University
Graduating: December 2026
GPA: 3.7
Location: Houston, TX
Focus: Agentic AI

Featured Projects

Production Systems

Use the cards to scan what shipped, what it does, and the stack behind it.

01 / Featured

Anna - Agentic Placement Automation System

A production AI agent built to automate contractor placement operations at scale. Processing over 1,000 documents per week, Anna autonomously stages candidates through the placement pipeline - work that previously required continuous manual effort from the recruiting team. The system reduced manual workload by approximately 90%, freeing staff to focus on higher-value decisions.

n8n
Claude API
Bullhorn CRM
Twilio
Gmail
Supabase
PostgreSQL
Microsoft SQL Server

Internal company project

02 / Featured

Oil AI - Drilling-Success Screening Platform

A geospatial ML platform that screens historical drilling success across two US basins - Colorado's DJ Basin and the Texas Eagle Ford - built end-to-end on public data: 124,302 wells and 15.8 million production records. Point-in-time-safe modeling with 15 km spatial-block cross-validation and isotonic calibration; LightGBM beat a strict drilling-density baseline at 0.842 vs 0.753 AUC (0.875 on Eagle Ford gas), backed by SHAP explainability, drift monitoring, an adversarial methodology audit, and a 78-test suite. Ships as an interactive MapLibre globe terminal with a guarded natural-language AI analyst that answers questions and flies the map.

Python
LightGBM
DuckDB
FastAPI
MapLibre GL
deck.gl

Live DemoPublic data / $0 infra

03 / Featured

MOVE - Motor Output vs. Visualized Execution

An end-to-end 64-channel motor-imagery EEG pipeline built for 4-class intention decoding across 86 subjects and 7,000+ trials. Applied subject-independent GroupKFold validation with EEGNet and Riemannian ML, achieving 42-45% balanced accuracy against a 25% chance baseline. Deployed as a real-time inference API with confidence scoring, artifact detection, and ABSTAIN safety gating for interpretable BCI output.

Python
MNE-Python
TensorFlow / Keras
Scikit-learn
FastAPI

GitHub

04 / Featured

System Design Walkthroughs

Three interactive walkthroughs of hard distributed-systems problems - a globally distributed key-value store, planet-scale video streaming, and real-time ride dispatch. Each states the problem, then builds the answer step by step: requirements, capacity arithmetic, the naive design, the point where it breaks, and the fix - all on a live 3D board you can reach into and rearrange. Drag a storage node around the hash ring and the key space re-divides under your hand; every animation exists to show cause and effect, not decoration.

Three.js
Vanilla JS
Consistent Hashing
CDN Economics
Geospatial Indexing

Open Walkthroughs

More Projects

Also Built

Quant Research

REGIME - Honest Trading-Model Evaluation

A Markov regime-switching trading model wrapped in a simulation-only evaluation harness built to answer one question with data: does the model beat the dumbest baseline in finance? It doesn't - and the harness proves it, showing exactly zero edge over persistence across 20 markets and 16,773 held-out predictions, verified pointwise, with positive controls in CI confirming the harness detects real switch structure when it exists. A negative result, reported honestly, with the receipts.

Python
statsmodels
Kalman Filter
pandas
pytest / CI

Live SiteSimulation only

Systems

Multi-Agent Task Orchestrator

A human-in-the-loop multi-agent system where seven specialized AI agents collaborate under a central orchestrator to plan, execute, and check complex multi-step work. Confidence scoring and risk flagging route high-risk steps to human approval gates, and a persistent memory layer spanning five learning types lets the system adapt from corrections instead of repeating mistakes.

FastAPI
PostgreSQL
React
TypeScript
Claude API

Platform

Admission Copilot

AI-powered university program discovery and application management platform. Personalized program matching, task-list generation, and guided planning workflows backed by a scalable Prisma + PostgreSQL data layer.

Next.js
TypeScript
Tailwind CSS
Prisma
PostgreSQL
Gemini API

GitHub

NVIDIA GTC Hackathon / Finalist

AI Agent Security Sandbox

Adversarial evaluation framework for browser-based AI agents. Tests against 40+ prompt-injection and unsafe-action patterns with reproducible evidence logging, context-aware risk scoring, and remediation reporting.

Python
Playwright
Static Analysis
NVIDIA Nemotron

Experience

Work History

North American University

Automation Engineer

July 2026 - Present

Building internal automation systems powered by locally hosted large language models that process and automate workflows across the university's departments. Administers the university's Salesforce org (TargetX) - Flows, validation rules, record-level security, and REST API integrations - and operates AI-driven n8n workflow agents connecting Salesforce, email, and document systems for automated application screening and routing.

VC5 Partners

AI Engineer Intern

February 2026 - July 2026

Built AI systems for legal document review, recruiter onboarding, and accounting automation. Architected a 3-agent document pipeline (extractor, reviewer, assembler) for oil & gas legal contracts with Claude, FastAPI, Azure Document Intelligence, and pgvector; a 4-stage agentic n8n placement pipeline across Bullhorn, Magnit, Twilio, and PostgreSQL that eliminated most back-office manual work; a QuickBooks Desktop integration driven by Claude agents through Flask middleware and a Cloudflare tunnel; and a GitHub scraper with LLM enrichment processing 1,000+ candidate profiles per week.

North American University

Salesforce Automation Engineer (Student)

September 2023 - February 2026

Built an AI-assisted admissions triage system on Salesforce (Flows, validation rules, REST integrations) that classified applications, flagged exceptions, and routed records autonomously - reducing processing time by 32% and manual errors by 41% across 3 admissions cycles. Deployed 8 production n8n workflow agents connecting Salesforce, email, and document systems, saving 40+ staff hours per cycle across 700+ applications.

Awards & Certifications

Verified Signals

Datathon Winner / 1st Place

Rice University

Hackathon Finalist

NVIDIA GTC Hackathon / AI Agent Security Sandbox

Generative AI LLMs

NVIDIA Certified Professional / NCP-GENL

Accelerated Data Science

NVIDIA Certified Professional / NCP-ADS

Currently

Technical Focus

Agentic systems in production

Orchestration, grounding, and the deterministic guardrails that keep multi-agent pipelines safe enough to run unattended against real business data.

Interpretable ML for decision support

Confidence scoring, abstain behavior, and artifact detection - so a model that is unsure says so instead of guessing.

Graduate research

Pursuing graduate research in AI systems, with a focus on the reliability and evaluation of autonomous agents.

Open to opportunities

Let's Connect

I'm currently open to internship opportunities, research collaborations, and PhD program connections in AI and machine learning. If you're working on something interesting, reach out.

Email Me LinkedIn GitHub Resume