Daniel Silva, ingeniero de software
Available now for full-time remote roles Caracas, Venezuela · 100% remote · UTC−4

Daniel SilvaSoftware that keeps running when the network doesn't.

Senior Backend Engineer · Full-Stack · Real-Time Systems

I design and ship the software industrial operations depend on: real-time telemetry, networked hardware control, and systems that keep working when the network does not. Five years, four production systems, ~3,000 commits.

Easy to hire: as an independent contractor (W-8BEN) or through an EOR such as Deel. No local entity needed.

  • 5+Years shipping to production
  • 4Systems in production
  • 3K+Commits over 5 years
  • 100%Remote · LATAM ⇄ NA/EU

Why hire Daniel Silva?

Measurable results in production systems, not demos.

1–10 ms

Real-time latency

Replaced HTTP polling with a full-duplex WebSocket architecture: from 200–500 ms down to 1–10 ms and up to −95% bandwidth.

5 yrs

Running unattended

A networked industrial hardware-control system, designed and maintained as principal author, in real field use with no technical support on site.

100%

Offline operation

Offline-first ecosystem with bidirectional cloud↔local sync: the plant works with no internet and reconciles when it returns.

0 → prod

End-to-end product

Built and run DropAudio CCS, a live e-commerce platform: Nuxt 3 + Supabase/PostgreSQL, multi-currency checkout and 120+ verified reviews.

Case studies

Four systems, told as problem, solution and result.

My own product, in production

DropAudio CCS

120+verified reviews from real customers

The problem
An audio distributor in Venezuela sold through WhatsApp and spreadsheets: no browsable catalog, prices in three currencies that change daily, and deliveries coordinated by hand.
What I built
A single Nuxt 3 and Supabase app covering the public store and the admin panel: a 3-step headphone recommender, a technical comparator, checkout in USDT, Zinli and Pago Móvil with live BCV rates, inventory, per-km deliveries and push alerts.
The result
The store runs every day. If Supabase goes down, the backup catalog keeps selling and checkout still builds the order over WhatsApp.
  • 19 models in the catalog
  • Writes locked down with RLS and is_admin()
  • Delivery reminders via pg_cron and Web Push
  • Nuxt 3
  • Supabase
  • PostgreSQL
  • Tailwind
  • Vercel
  • Web Push
dropaudio.mp4

Internal tool with local AI

DropAudio Reel Studio

2 h15 minof editing per reel

The problem
Every reel for the store meant manually reviewing all the phone footage to find usable takes. About two hours per piece.
What I built
A desktop app that transcribes with faster-whisper, describes and scores each take with vision models running on the same machine, writes the script and delivers the assembled 1080×1920 CapCut project.
The result
Editing drops to about 15 minutes, mostly unattended. Footage never leaves the computer.
  • Static frames skipped before any model call
  • Resumable analysis: a crash costs one file, not the run
  • Rule-based script fallback if the model fails
  • Python
  • faster-whisper
  • Ollama
  • pyCapCut
  • ffmpeg
  • PyInstaller
View the code
reel-studio.mp4

Industrial system for a client

Industrial monitoring with networked camera control

5 yearsrunning on the plant floor, with no on-site tech support

The problem
Operating industrial cameras split across two networked machines, somewhere nobody can go restart anything if it hangs.
What I built
A Python desktop app where the same executable takes its role on each machine with no manual setup, transfers images in chunks and puts an explicit timeout on every network boundary.
The result
A powered-off camera no longer freezes the interface. The system has been in real use for five years and ships with a versioned installer.
  • Automatic role at startup: server or client
  • Chunked image transfer with reassembly
  • Automated tests with pytest
  • Python
  • PyQt5
  • OpenCV
  • TCP
  • pytest
monitoring.mp4

Applied research

Video stabilization

2 enfoquesclassic and deep learning, comparable side by side

The problem
Handheld phone video shakes. My undergraduate thesis proposed AI-predicted adaptive weights, but the original code only ran from the command line and did not implement them.
What I built
Two stabilizers: one using a mesh, optical flow and interpolation, with a network predicting its weights and a before/after web UI; and one that estimates camera motion with a distilled PWC-Net and smooths the path with quadratic programming.
The result
Results are measured with MSE, RMSE, PSNR and SSIM, and the GPU version stabilizes new videos without retraining.
  • Live progress over SSE from FastAPI
  • Geometric and photometric correction combined
  • Undergraduate thesis, University of Zulia
  • Python
  • PyTorch
  • OpenCV
  • scikit-learn
  • FastAPI
  • CUDA
stabilization.mp4

Systems in production

Four systems built in parallel over five years. Client work under confidentiality: I describe architecture and decisions, never business data.

industrial_iot_hardware_control.md2022 — 2026 · 5 years

Industrial IoT hardware control

Principal author and architect

Desktop application for coordinated control of networked industrial hardware, built to run with no technical support present in the field.

  • Principal author
  • 5 years in real use
  • Versioned installer
  • Zero manual setup by the operator and recovery from network drops: built to run unattended.
  • Explicit timeouts at every network boundary so a disconnected machine never freezes the UI.
  • Smooth, interactive visualization of live data embedded in the desktop app.
  • Refactored from monolith to modular library without stopping delivery, with automated tests and a versioned installer.
PythonPyQt5OpenCVSocket.IOmatplotlibpytestPyInstaller
industrial_web_platform.md2022 — 2026 · 4 years

Industrial web platform

Created the repository · integration lead

Full inspection and analytics platform for industrial assets: from the initial repository "chassis" to a product in production, integrating the work of a ~14-person team.

  • ~2,000 commits
  • ~14 contributors integrated
  • 11+ integration branches
  • Owned the entire stack: Nuxt/Vue frontend, Express/TypeScript backend, Sequelize/MySQL data layer and AWS deployment.
  • Crack-analytics engine and 2D/3D visualization (Highcharts, D3, Three.js) over industrial asset geometry.
  • Real-time voice chatbot: WebSocket, audio capture and synthesis, speaker detection and live mode with auto-recovery.
  • Owned the feature→production cycle: continuous versioning, query performance testing, build and deploy.
Nuxt/VueTypeScriptExpressMySQLThree.jsWebSocketsAWS
offline_first_ecosystem.md2023 — 2026 · 3 years

Offline-first ecosystem

Principal author since day one

Desktop client that runs 100% offline on plant floors and syncs bidirectionally with the cloud whenever connectivity returns.

  • ~250 commits
  • ~845 files
  • Self-contained installers
  • Cloud↔local sync with per-endpoint state, connectivity detection and retries: zero lost work in the field.
  • Python orchestrator that brings up a portable database, Node/TypeScript services and the API, all packaged in an installer that resolves its own dependencies.
  • Custom AI audio transcription and synthesis pipeline (FastAPI/Uvicorn) integrated into the product.
  • CI/CD with GitHub Actions, tests with configuration isolation and database connection pooling.
PythonFastAPINode/TSMariaDBS3GitHub ActionsPyInstaller
industrial_time_series_analysis.md2024 — 2026 · 2 years

Industrial time-series analysis

Principal author

Tool that automatically detects thermal cycles and key gradient points in sensor series, replacing manual spreadsheet analysis.

  • CLI-parameterized pipeline
  • JSON · Excel · PNG outputs
  • Rust port exploration
  • Peak and valley detection with SciPy over a smoothed signal, plus interpolation of faulty readings so detection never breaks.
  • Polars-based processing, with the pipeline adapted to two source-format changes without a rewrite.
  • Engineering-ready outputs: structured JSON per cycle, an Excel table and an annotated plot.
  • Documented Rust rewrite exploration (calamine + polars), archived as reference.
PythonPolarsNumPySciPymatplotlibRust

Private repositories (client code / NDA). I can walk through architecture, decisions and trade-offs on a call.

Where has Daniel Silva worked?

Five years shipping software to production for industrial clients in Canada and the maritime/oil sector.

  1. Lead Performance & Systems Engineer

    Ea2technology · Canada · Remote · Full-time

    Jul 2021 — Sep 2026
    • Designed and deployed an industrial Digital Twin telemetry system for real-time structural monitoring, processing high-frequency sensor data to anticipate critical failures.
    • Designed a low-latency, full-duplex WebSocket architecture that replaced HTTP polling: from 200–500 ms to 1–10 ms and up to −95% bandwidth.
    • Built a desktop application for coordinated remote control of industrial IoT hardware across networked machines, designed for zero-supervision operation under unreliable networks.
    • Built an offline-first ecosystem with bidirectional sync, enabling 100%-offline operation and automatic reconciliation once connectivity is restored.
    • Integrated an AI-based audio transcription and synthesis pipeline into the platform.
    • Led the migration of legacy monoliths to cloud-native microservices on AWS, optimizing compute-intensive pipelines with C++ and Rust.
    RustC++AWSWebSocketsIoTOffline-first
  2. Senior Software Engineer (Contract)

    VIC · Vessel Inspection · Remote · Maritime / Oil & Gas

    Jan 2023 — Jul 2025
    • Built and containerized backend services with NestJS and Docker, with CI/CD pipelines on GitLab CI, GitHub Actions and AWS CodePipeline.
    • Audited 4 stacks and found 3 memory leaks and concurrency bottlenecks in point-cloud processing (up to 1.2M points): a sequential job went from 12–14 h to 5 h and data navigation from ~1 s to 15–20 ms.
    • High-performance video stabilization engines with optical flow and matrix interpolation (Python, NumPy, SciPy, OpenCV).
    • Global motion estimation with FlowNet (distilled deep architectures), combining classical math with machine learning.
    • Time-series pipelines that detect cycles and inflection points in industrial sensor data (Polars, SciPy), configurable via CLI.
    NestJSDockerPythonOpenCVPyTorchRust
  3. IT Support Specialist

    RenéDessés de Venezuela · Caracas, Venezuela · Full-time

    Jan — Sep 2021
    • Maintained equipment, software and Apache/Linux servers.
    • Installed and administered enterprise networks, including Cisco infrastructure.
    LinuxApacheCisco

What has Daniel Silva built?

A curated selection of my work on GitHub.

dropaudio_ccs.vue Live

DropAudio CCS

Complete e-commerce in production with an admin panel: catalog, multi-currency checkout, audio recommender and real-time deliveries.

120+ verified reviews · checkout in three currencies

system_stabilization_interpolation.py Public

System Stabilization Interpolation

Video stabilization system for mobile devices using motion meshes and interpolation — undergraduate thesis.

Results measured with MSE · PSNR · SSIM

flownet_video_stabilization.py Public

FlowNet Video Stabilization

Deep learning video stabilization: global motion estimation distilled from optical flow.

The GPU version stabilizes new videos without retraining

camara_ocr_python.py Private

Camara OCR Python

Computer vision and OCR tool to assist industrial inspections.

5 years running on the plant floor, with no on-site support

dropaudio_reel_studio.exe Public

dropaudio reel studio

Desktop app that turns phone footage into a finished CapCut reel draft: transcribes, scores every take, writes the script and assembles the project. From ~2 h of manual editing to ~15 min.

All local: footage never leaves your computer

realtimevoiceassistant.exe Public

RealtimeVoiceAssistant

Real-time conversational voice assistant: transcription, NLP and speech synthesis with Gemini Live.

Critical audio in Rust, sub-100 ms responses

extract_rif.sh Public

extract rif

HTTP microservice extracting structured RIF (SENIAT) data from PDFs or images with a dual-engine OCR pipeline.

Always-valid response, with a fallback engine

bolsa_valores_caracas_api.sh Private

bolsa valores caracas api

Public FastAPI serving data on stocks listed on the Caracas Stock Exchange.

GET /acciones · GET /acciones/{simbolo}

Education & Certifications

University degree completed in parallel with professional experience.

education.db

Education

  • Bachelor's Degree in Computer ScienceLUZ-IUTA · University of ZuliaGraduated Apr 2025
  • Associate Degree (TSU) in ITLUZ-IUTA · University of Zulia2018 — 2021

Languages

Spanish (native) · English B2 professional

certifications.db

Certifications

  • Vue.js 2 Professional — Platzi (2024)
  • TypeScript: Advanced Types & Functions — Platzi (2024)
  • TypeScript Fundamentals — Platzi (2024)
  • Web Optimization — Platzi (2024)
  • Python: NumPy & Pandas — Platzi (2024)
  • Introduction to Generative AI — Duke University / Coursera (2025)

Which technologies does Daniel Silva work with?

Tools I use every day.

Python TypeScript Rust Vue 3 Nuxt 3 React Native Vite PyTorch OpenCV Python TypeScript Rust Vue 3 Nuxt 3 React Native Vite PyTorch OpenCV
contact.md

Hiring a backend or full-stack engineer?

I reply within 24 hours. Happy to do a technical screen or walk you through any of these systems on a call.

Professional references available on request.

Easy to hire: as an independent contractor (W-8BEN) or through an EOR such as Deel. No local entity needed.