OPEN TO SELECT ENGINEERING ROLES
DEVIN HONSON · FOUNDER, CEO & LEAD ENGINEER
I build systems
where milliseconds matter.
Real-time media engines, GPU pipelines, ML audio, streaming infrastructure, backend systems, and product layers designed to survive production not just demos.
7+
YEARS BUILDING
72 → 4.82
MS ENCODER LATENCY
GPU · ML · MEDIA
END-TO-END OWNERSHIP
FOCUS
REAL-TIME
MODE
BUILD / SHIP
RUST
DIRECT3D
TAURI
PYTHON
DJANGO
FFMPEG
WEBSOCKETS
NGINX
ABOUT / FOUNDER ENGINEER
I started Real Labs as a personal experiment in real-time media engineering. Seven years later, it became a production engine powering multiple platforms. I build systems that survive production, not just demos.
01 // WORK STATUS
Current mission &
systems shipped.
I do not separate “product work” from “systems work.” The product is only as good as the capture path, runtime state, queueing discipline, observability, deployment, and failure recovery underneath it.
CURRENT ROLE
ACTIVE
Founder, CEO & Lead Engineer
Real Labs Development LLC
Own product architecture, native media systems, GPU/runtime engineering, ML audio, web/backend, deployment infrastructure, operations and technical strategy.
OWNERSHIPProduct → engine → infra
CORE DOMAINSGPU / Media / ML / Platform
OPERATING STYLEInstrument · isolate · optimize
STATUSProduction + launch-ready
DELIVERY METRICS // PUBLIC-SAFE
Proof of throughout without exposing private client details.
7 proprietary engines shipped · 3 production modules · 12 active RVC projects live · 916 Rust/audio tests passing
CAREER // CURRENT + COMPLETED
Professional engineering engagements.
CURRENT
FOUNDER
Founder, CEO & Lead Engineer
Real Labs Development LLC
2019 — PresentOwn product architecture, native media systems, GPU/runtime engineering, ML audio, web/backend, deployment infrastructure, operations and technical strategy.
- Built four proprietary engines with zero contractors
- Production-hardened RealVideoCreator across Django, Python, FFmpeg and dedicated services
- Built LUNA real-time GPU compositor and A/V runtime in Rust + Direct3D 11 + WASAPI
- Built VoiceLab in-house inference workflows and Real Gamers native streaming foundation
RustPythonDjangoPostgreSQLDirect3D 11DXGIWASAPIFFmpegRedisCeleryNginxsystemd
SHIPPED SYSTEM
01
LUNA — Real-Time Graphics & Media Runtime
Founder / Systems Engineer / GPU Runtime ArchitectProprietary real-time GPU compositing, overlay, A/V synchronization and media runtime powering the Real Labs ecosystem.
RustDirect3D 11DXGIWASAPIFFmpegH.264AACMKV
SHIPPED SYSTEM
02
RealVideoCreator
Founder / Full-Stack & Media Pipeline EngineerProfessional video creation platform production-hardened across web, database, media processing, billing, staffmail and dedicated service layers.
Django 5.1.7Python 3.12PostgreSQL 16FFmpeg 6.1GunicornDaphnesystemdNginx
ACTIVE SYSTEM
03
Real Gamers Streaming Studio
Founder / Native Media & Streaming EngineerNative Windows media engine and streaming foundation built around low-latency capture, Direct3D compositing, audio timing, scenes, overlays and platform integration.
RustDirect3D 11DXGIWASAPIOAuth PKCEWebSocketsKick
SHIPPED SYSTEM
04
VoiceLab In-House Voice Engine
Founder / ML Audio EngineerIn-house voice cloning and generation system with local inference, dataset workflows, training/export tooling and production service isolation.
PythonChatterbox TurboCPU inferenceML audiodataset toolingsystemd
COMPLETED JOBS // SYSTEMS DELIVERED
- RealVideoCreator production platformDjango, PostgreSQL, FFmpeg, billing, staffmail, export services.
- LUNA Engine maturity milestoneRust, D3D11, WASAPI, MKV/MP4 media pipeline, overlay and scene runtime.
- VoiceLab production-ready engineIn-house voice generation and CPU/GPU inference workflows.
- Real Gamers streaming foundationNative Windows media engine, Direct3D compositor, Checkpoint010 complete.
CURRENT JOB // ACTIVE BUILD
Checkpoint011 + public launch infrastructure
Expanding Real Gamers toward RTMP transport, public beta readiness, creator onboarding, API usage paths, studio-seat workflows, and infrastructure scaling.
02 // TECHNICAL DEPTH
I understand the system
under the feature.
My strongest work happens when the bug crosses subsystem boundaries: GPU → CPU, capture → presentation, inference → audio timing, provider event → product state, service → deployment.
Frame ownership & timing
Reasoning about frame ownership, validity, queue residency and the synchronization boundary actually creating latency.
- capture freshness
- live-edge behavior
- swapchain lifecycle
- present discipline
GPU resource lifecycle
Allocation, invalidation, device-bound state, identity-aware rebuild logic and safe GPU/CPU ownership boundaries.
- resource identity
- reallocation
- device-bound state
- GPU/CPU boundaries
Latency decomposition
Instrumenting the complete path so queueing, blocking, submission, encode and delivery costs are measured separately.
- queue depth
- blocking points
- stage timing
- end-to-end budgets
Real-time ML scheduling
Treating inference as a deadline-driven live system where continuity, backpressure and worker behavior matter as much as throughput.
- GPU workers
- CPU fallback
- continuous speech
- backpressure
State authority
Defining explicit ownership across distributed clients, normalized provider events, reconnect behavior and scene authority.
- event envelopes
- scene authority
- reconnect semantics
- state normalization
Production operations
Shipping with TLS, reverse proxies, isolated workers, health checks, Linux services, databases, backup and rollback discipline.
- Nginx / systemd
- Linux services
- PostgreSQL
- deployment / rollback
COLLABORATION + DELIVERY
Production engineering includes the handoff.
I build so another engineer can understand the system: version-controlled code, deploy scripts, documented service boundaries, environment separation, and observable runtime behavior.
CASE STUDY // ENCODER PIPELINE
72ms
→
4.82ms
93.3% latency reduction
The encoder was not the diagnosis. The full pipeline was.
- ProblemEnd-to-end encoder latency was too high for responsive live production.
- InvestigateInstrumented queueing, synchronization, frame ownership, submission and encode stages separately.
- FixRemoved avoidable blocking and tightened frame-flow discipline across the path.
- ResultReduced observed encoder-path latency from 72 ms to 4.82 ms.
03 // ARCHITECTURE
One ecosystem.
Multiple system boundaries.
The value is not a pile of frameworks. It is knowing what each layer owns, what it must never own, and how data, media, state, and failures move between them.
Native engine boundary
Time-sensitive capture, GPU, audio, and encoding work stays close to the hardware and native runtime.
Service boundary
Long-running jobs, ML workers, exports, APIs, and operational services are isolated by responsibility.
Product boundary
The UI communicates intent. It should not secretly become the owner of media timing or infrastructure state.
Observability boundary
Collectors report what the system is doing so recovery decisions can be based on evidence instead of assumptions.
INPUTCapture / Audio / Events
›
ENGINELUNA Runtime
›
SERVICESML / API / Export
›
PRODUCTStudio / Web / Stream
COLLECTORS
HEALTH GOVERNOR
DEPLOYMENT
SECURITY
04 // WHY REAL LABS EXISTS
I kept building until the project became a platform.
I started Real Labs as a personal experiment in real-time media engineering. Seven years later, it became a production engine powering multiple platforms. I build systems that survive production, not just demos.
Once the foundation became mature, new systems could share the same thinking: explicit ownership, measurable timing, isolated services, observable health, recoverable state, and production-grade deployment.
That is the kind of engineering role I am looking for: a place where technical depth matters, where I can own difficult problems, and where shipping the system means understanding the whole path.