hawari.tech
Book a technical call
Menu
Practice areas Selected proof About Basil Engagement models Book a technical call
independent consultancy · principal-led

Edge AI, Computer Vision, and Distributed Systems Engineering.

HawariTech is an independent consultancy led by Basil Hawari. I help engineering teams design, optimize, and deploy real-time perception models and scalable cloud architectures — from resource-constrained silicon to production backends.

Schedule a Technical Call ↗ Email Directly
focus
Edge AI · Computer Vision · Embedded MCUs/MPUs · AWS Cloud
target silicon
NVIDIA Jetson · STM32N6 · ARM architectures
location
Remote worldwide · Based in Haifa, Israel
15+
years shipping production systems across the full stack
90%+
inference latency reduction on a Jetson safety deployment
20K+
daily users served on platforms architected & scaled
99.9%
uptime on mission-critical infrastructure under management
practice areas

I group work by the technical problem you are trying to solve.

Not a bare list of technologies — three problem spaces where model accuracy, hardware constraints, and latency budgets collide with real deployments.

P—01

Edge AI & Computer Vision Deployment

⚠ the problem

Models running in Python on high-end GPUs fail under physical deployment constraints — thermal throttling, strict latency budgets, and memory limits.

  • Custom inference engines in C++ on TensorRT, ONNX Runtime, and CUDA.
  • Model optimization, quantization, and pruning for real-time edge execution.
  • Fusion of classical vision (OpenCV, homography, multi-view geometry, optical flow) with deep networks for tracking and segmentation.
  • Integration from NVIDIA Jetson to next-gen NPUs like the STM32N6.
c++tensorrtonnx cudaopencvquantization jetsonstm32n6
P—02

Real-Time Media & IoT Architecture

⚠ the problem

Edge devices must stream video, communicate status, and receive commands without dropping frames or introducing unacceptable latency.

  • Ultra-low-latency video streaming pipelines with WebRTC, GStreamer, and C++.
  • Reliable IoT telemetry, command, and control backends via AWS IoT Core and MQTT.
  • Edge-to-cloud sync for automated synthetic data capture, metrics logging, and model monitoring.
webrtcgstreamermqtt aws-iot-corec++
P—03

Full-System Prototyping & Production Hardening

⚠ the problem

Startups and R&D groups need an end-to-end working system to validate a product concept or unblock a hardware roadmap.

  • Rapid software prototyping on pre-release silicon and custom hardware rigs.
  • Computer vision engines integrated into native mobile apps — iOS, Android, wearables.
  • Serverless cloud backends (AWS Lambda, AppSync, DynamoDB) sized for predictable cost and near-zero maintenance.
aws-lambdaappsyncdynamodb flutteriosandroid
selected proof points

Outcomes engineers hire on — not marketing claims.

Structured problem → action → outcome summaries. Details beyond what is NDA-safe are shared on a technical call.

Real-Time Edge Safety Vision

Coral Smart Pools — edge-AI R&D lead
nvidia jetsonc++tensorrt webrtcquantization
◈ context

Commercial pool safety system requiring continuous, sub-second human detection under fixed thermal budgets.

⚒ implementation

Replaced standard inference code with a tailored C++ pipeline on NVIDIA Jetson using TensorRT and custom ResNet optimizations; built low-latency WebRTC feed delivery.

✓ outcome

Reduced inference latency by over 90% while running within fixed thermal limits.

Discuss this project under NDA →

Edge NPU Early-Adopter Prototype

STMicroelectronics STM32N6 — pre-release silicon program
stm32n6neural-art npuc/c++ armquantized models
◈ context

Pre-release silicon integration for next-generation edge hardware, before public tooling and silicon availability.

⚒ implementation

Architected firmware and software layers targeting the STM32N6 Neural-ART NPU, evaluating accuracy trade-offs of quantized models against the on-chip compute budget.

✓ outcome

Delivered an operational prototype ahead of official silicon availability, establishing feasibility for low-cost hardware product lines.

Discuss this project under NDA →

Scalable Serverless Platform

FIGZI — co-founder, end-to-end technical lead
awsappsync · graphqllambda dynamodbflutter
◈ context

Event-driven mobile platform with high-concurrency location queries from launch day.

⚒ implementation

Built an event-driven backend on AWS — AppSync (GraphQL), Lambda, and Amplify — paired with a Flutter client.

✓ outcome

Supported 10,000+ users in year one with minimal infrastructure overhead and zero dedicated DevOps personnel.

Discuss this project under NDA →
about basil hawari

A principal engineer you work with directly.

I am a Principal-level software and AI engineer with 15+ years designing and shipping systems across the entire stack — silicon-level optimization, low-latency C++ services, production cloud infrastructure, and mobile applications.

Prior to founding HawariTech, I led R&D initiatives for safety-critical edge-AI hardware (commercial pool-safety systems on NVIDIA Jetson), built C++ robotics and IoT control software integrated with AWS IoT and MQTT, and co-founded a venture-backed startup, architecting its serverless backend from zero. Earlier, I led a DevOps transformation that moved a mission-critical platform from weekly to multiple daily deployments at 99.9% uptime, and scaled an LMS to 20,000+ daily users.

That career spans every layer of the compute continuum — from a neural network quantized for an ARM Cortex-M class NPU, through Jetson edge nodes running optimized inference, to event-driven cloud backends orchestrating the fleet. Few consultancies can design across that whole spectrum, because few have personally built at every layer.

When you hire HawariTech, you work directly with me. There are no account managers, outsourced junior developers, or handoff layers. I write the architecture, profile the bottlenecks, and deliver the production code.

how we work

Three ways to engage — pick the entry that fits.

Each model starts with a technical discussion, not a sales deck. Scope and estimates are grounded in your codebase and hardware, not generic proposals.

E—01
fixed scope · 1–2 weeks

Architecture & Feasibility Audit

goal

Determine whether a proposed edge-AI, vision, or cloud architecture is viable before you commit significant hardware or hiring budget.

deliverables

Detailed technical audit · silicon selection analysis · latency/accuracy profiling on sample data · actionable implementation blueprint.

Start with an audit call →
E—02
project-based · 4–8 weeks

Prototype & Systems Delivery

goal

Build an operational proof of concept, or ship a specific subsystem — a custom inference pipeline, WebRTC streaming engine, or IoT backend.

deliverables

Documented, tested C++/Python/cloud codebase · integration tests · deployment scripts · performance benchmarks.

Discuss your subsystem →
E—03
retainer · embedded

Fractional Principal Engineer

goal

Strategic technical leadership and hands-on implementation for teams that lack senior embedded-AI or systems-architecture expertise.

deliverables

System architecture design · sprint-level code contributions · code reviews · technical mentorship for your internal team.

Talk about embedding me →

Bring the hard problem first.

No intake forms, no account managers, no qualification funnel. If your problem involves real-time perception, constrained silicon, or a backend that must not fall over — send the context and the constraints, and let's talk architecture.

logistics
  • Based in Haifa, Israel · remote worldwide
  • Working with teams in North America, Europe, and Israel
  • Engagements from 1–2 week audits to embedded retainers
  • NDA-friendly — signed before any architecture detail is shared
handshake_protocolready
step 1email / call — context + constraints
step 230-min technical discussion
step 3technical brief with honest estimates
no_sales_deck_required = true