Devon Clemente

    AI Process Automation Specialist

    Building intelligent automation solutions that streamline workflows and eliminate inefficiencies

    📍 Greater NY/NJ metropolitan area

    Scroll to learn more
    About Devon

    Transforming Operations ThroughAI Automation

    My Story

    Nine years in operations taught me one thing: most business problems are process problems. I've managed $20M+ in inventory, run crews across four states, and cut delivery failures by 75% — all by finding where the system breaks and fixing it.

    Now I do the same thing with AI. After earning my AI Automation certification, I build automated workflows that eliminate the manual bottlenecks I used to fight every day. Make.com, UiPath, API integrations — these are just better tools for the same job I've always done: making operations actually work.

    Most people automating business processes have never been inside a broken one. I've spent nine years there. That's the difference.

    98% On-Time Delivery

    First-responder supply chain at Air Brake & Equipment

    25% Business Growth

    Client acquisition and operational scaling at Maverick

    50+ Crews Managed

    Multi-state installation teams across the tri-state area

    7+ Automation Projects

    End-to-end workflows built with Make, UiPath, and AI

    Core Competencies

    Process Automation
    AI Implementation
    Operations Management
    Workflow Optimization
    Business Intelligence
    Performance Analytics

    Automation & AI Skills

    Make
    Zapier
    N8N
    Miro
    Nebius Studio
    LLMs
    API Integration
    RPA Development
    UiPath
    GitHub
    VS Code
    Live Benchmark Data

    Cloud vs LocalLLM Benchmark

    We ran 4 AI models on the same real ops data — summarization, priority extraction, JSON output, and message drafting. Same prompt. Every model. Here's what came back.

    Task: summarize
    Same input. Every model.

    Write a 3-sentence summary of what needs attention this week.

    Open items for the week:
    - Follow up with client on proposal sent Tuesday
    - Review Q1 budget draft before Thursday meeting
    - Respond to 3 support tickets marked urgent
    - Update project timeline — milestone 2 slipped by 4 days
    - Schedule onboarding call for new team member starting Monday

    Claude Sonnet

    Cloud
    2.4s47.2 tok/s

    Run it yourself:

    python3 llm-benchmark.py --tasks summarize

    Accurate and well-structured. For sensitive operational data, it's a non-starter — and even when the data probably isn't sensitive, we'd rather not find out the hard way.

    Claude Haiku

    Cloud
    1.1s68.4 tok/s

    Run it yourself:

    python3 llm-benchmark.py --tasks summarize

    Fast and cheap. Output quality held up on simple tasks but drifted on the more complex ones.

    âś“ DEPLOYED IN PRODUCTION

    Gemma

    Local
    4.2s28.6 tok/s

    Run it yourself:

    python3 llm-benchmark.py --models gemma --tasks summarize --skip-claude

    Most consistent across all four task types. Ran on a $1,100 Mac Mini — no cloud, no subscription, no data leaving the building. This is the one we deployed.

    Mistral

    Local
    6.1s19.4 tok/s

    Run it yourself:

    python3 llm-benchmark.py --models mistral --tasks summarize --skip-claude

    Capable model — output quality is version and size dependent. The 7B we ran was slower and less consistent than Gemma on structured tasks.

    Full results across all 4 tasks — summarize, priorities, classify, draft — in the sample report and article series below.

    Interactive Projects

    Games &Interactive Projects

    Interactive experiences built with modern web technologies. Combining creativity with technical skills to create engaging digital games.

    Roddy's StarTrack preview

    Roddy's StarTrack

    AI-First Development

    Created for my nephew Roddy, this digital adaptation brings a beloved physical board game to life. Built using an AI-first development approach: parallel PRDs generated by Claude and ChatGPT, both prototyped in Lovable, with the winning version refined in VS Code and deployed via Harness CI/CD pipeline.

    Technology Stack

    React
    TypeScript
    AI Integration

    Results

    Validated AI PRD methodology comparing Claude vs ChatGPT outputs. Established rapid prototyping pipeline using Lovable for quick iteration. Built production-ready game with comprehensive test suite via Harness.

    Development Process

    • •Generated competing PRDs with Claude and ChatGPT to test AI-assisted product design
    • •Prototyped both versions simultaneously in Lovable for rapid visual iteration
    • •Selected the stronger prototype and refined game logic, animations, and board layout in VS Code
    • •Deployed with Harness CI/CD pipeline including automated testing on every push
    Loading articles...