Engineer and researcher working across agentic AI, machine learning, quantum computing, and embedded systems.
Member of Technical Staff & SDE Intern at Amazon AGI Lab & Kiro Labs (May 2026 — August 2026)
B.S. Electrical & Computer Engineering (AI/ML), Physics, and Computer Science, with Distinction — Duke University (2023 — 2027)
My work sits where learned systems meet things that have to actually run — agents that are measured rather than demoed, quantum circuits benchmarked on real hardware rather than simulators alone, and models that end up on an aircraft or a wearable. Personal site and writing at mzoraiz.dev (source — mzoraiz-dev).
| Role | Org | When |
|---|---|---|
| Member of Technical Staff & SDE Intern | Amazon AGI Lab & Kiro Labs | May 2026 — present |
| AI Data Engineering Intern | Pfizer | Jan — Apr 2026 |
| Machine Learning Engineering Intern | Amazon Web Services | May — Aug 2025 |
| Software Engineering Intern | Amazon Web Services | May — Aug 2024 |
| Embedded Software Engineering Intern | Rockwell Automation | Jun — Aug 2023 |
| Quantum Computing Researcher | MIT Lincoln Laboratory | Feb — Dec 2022 |
| Software Engineering Intern | Rockwell Automation | Jun — Aug 2022 |
Amazon AGI Lab & Kiro Labs — architecting low-latency voice-to-agent workflows connecting Kiro code agents with streaming ASR/TTS infrastructure, and a real-time agent runtime for model-selectable coding that coordinates session state, streaming events, and autonomous code execution. (under NDA)
Pfizer — AI for genomics. Scalable data pipelines and ML workflows for model-driven analysis. (under NDA)
AWS · Agentic AI — built a serverless benchmark analysis platform adopted by the Kiro, Amazon Q, and AWS Agentic AI teams, automating large-scale code-synthesis evaluation and accelerating ML optimization by 40%. Orchestrated 120+ distributed service interactions across Lambda, DynamoDB, and S3, and integrated Amazon Bedrock to drive multi-armed bandit optimization. Work on cross-benchmark pattern detection, AST-based dependency mapping, and ML-driven error fingerprinting.
AWS · Amazon Q Developer — statistical analysis pipelines in Python, SQL and R; an LLM pipeline analyzer cutting error rates 15%; distributed LLM training optimization with SageMaker, Step Functions and Redshift for a 20% scalability gain.
Rockwell Automation — C++ firmware for L8Z motion controllers using RTOS scheduling and EtherCAT control loops (20% motion-accuracy gain), and automated C++ verification suites with GoogleTest and VectorCAST. Build pipelines on VMware, Docker and Jenkins cut build times 35%.
MIT Lincoln Laboratory — implemented the HHL algorithm in Q# and Qiskit for quantum fluid-flow and cognition simulations, engineering circuits with eigenvalue estimation, amplitude amplification and Hamiltonian simulation for 33% faster runtimes.
| Year | Venue | Work |
|---|---|---|
| 2026 | IEEE Quantum Week — QCE26, Poster Track | Quantum Bayesian Learner with Hardware-Aware Circuit Compression — how much entangling structure a quantum Bayesian learner can lose before it stops representing the task, validated on IBM's 156-qubit ibm_fez Heron device |
| 2024 | IEEE SIEDS — Systems and Information Engineering Design Symposium | Psycho Acoustic Testing to Determine the Optimal Frequency for Audible Safety Alerts for Freediving — with Garmin International Inc. and the Duke Marine Lab · code |
Posters — A Cyber-Physical UAV Framework for Agentic Precision Pollination through Integrated Computer Vision, Mission-Level Reasoning, and Embedded Mechatronic Actuation, Duke Pratt School of Engineering · poster + code
VOICE-EVPI — under submission to ICLR 2026.
Execution-grounded, cost-sensitive clarification for voice-driven coding agents.
When a coding agent hears you, it has to decide what to do about the parts it did not quite catch. Treating that decision as a cost-sensitive value-of-information problem — where the cost is a measured property of the code being changed — rather than a confidence threshold.
More soon.
Provenia Bio — a neuro-symbolic drug discovery system. In development.
Qadam · قدم — "a step forward." An accessible biotech venture that packs a whole prosthetics clinic into one portable case. Roughly 65 million people need a prosthetic or orthotic device and only about 1 in 10 has access to one — the barrier is rarely the limb itself, but the clinician, the lab, the power, and the supply chain around it. Qadam makes that whole apparatus portable: one ruggedized, solar-powered case carrying the fitting technology, tools, and standardized parts to custom-fit 15–20 limbs anywhere, plus the open IP to make more. The intelligence lives in the fitting process; the limb the patient keeps is robust, low-cost, and purely mechanical — nothing to charge, brick, or break. · live
Agentic AI and LLM systems
| Project | What it is |
|---|---|
| ombench | Memory and backtesting for operational agents — a bitemporal history substrate, a knowledge-base compiler, and a deterministic replay harness that measures whether compiled memory actually improves an agent on real historical tasks |
| targetONCO | End-to-end agentic precision oncology, from X-ray radiology to spatial proteomics tissue analysis, orchestrated by a single agent system |
| CtrlSlash | MCP-based documentation engine that auto-scrapes APIs and SDKs, powering semantic RAG search to cut AI IDE hallucinations — built on a $25k grant, with integrations for Cursor, Windsurf and Kiro (private) |
| agentic-pollination-uav | Autonomous pollination UAV — mission simulation and a real flight stack sharing one state machine, with an LLM planner, RAG mission memory, and a UCB1 bandit over detection thresholds |
Quantum
| Project | What it is |
|---|---|
| qb-learner-compression | Hardware-aware compression of a quantum Bayesian learner, benchmarked against post-transpile two-qubit gate count on real IBM hardware · QCE26 |
| DuQuantum | Duke quantum computing community site |
Health, bio, and genomics
| Project | What it is |
|---|---|
| LifeEdit-geneclassifier | ML classifier built for Life Edit Therapeutics detecting edited vs unedited cells from single-cell expression data |
| RevealGenomics | BRCA-gene cancer detection model, with Reveal Genomics |
| SonicSync | Submersible psychoacoustic wearable delivering audible safety alerts to freedivers, with Garmin · SIEDS 2024 |
Security, privacy, and signals
| Project | What it is |
|---|---|
| cipher-shield | Aggregate statistics on demographic data without ever decrypting a record — split-key homomorphic encryption with on-chain validation |
| SIDQ | Spoof identification under degraded quality |
Embedded and hardware
| Project | What it is |
|---|---|
| fpga-hologram | Persistence-of-vision rotating hologram on a Nexys A7-100T FPGA |
| atlas-loom | Open-source global exploration viewer |
Awards
- Amazon Future Engineer Scholar — 1 of 250 internationally, $40k, for excellence in engineering and computer science (2023)
- Lockheed Martin STEM Scholar — 1 of 100 nationally, $40k, for excellence in engineering and innovation (2023)
- Dominion Energy Educational Equity Scholar — 1 of 40 nationally, $40k, for national social impact (2023)
- Congressional Commendation — Congressman David Joyce, for community advancement in childhood education (2022)
- TreeHacks winner — CipherShield (2025)
Research and activities
- Dark Matter & AI Research, Kotwal Labs — high-energy physics, using AI and FPGA-integrated circuitry in dark matter search
- ProductSpace@Duke — Fellowship Director, leading technical product management education and mentoring multidisciplinary teams
- HackDuke — Organizer, coordinating logistics, sponsors and teams for tech-driven social impact
Languages
Machine learning and AI
Quantum
AWS
Cloud and infrastructure
Data
Embedded and hardware
Web and mobile
Design and CAD
Spoken — English, Urdu, Hindi (native) · Arabic, Spanish (professional working) · Turkish, ASL (basic)


