mattrguerra.comHouston area1 / 6
Matt Guerra
  1. 01AI & Automation
  2. 02Photography
  3. 03Design
  4. 04About
  5. 05Notes
  6. 06Contact
Section 01 · Twelve systems and projects

AI & Automation

Systems designed and built with AI-assisted development, and used every day. Shown here as rebuilds running on sample data.

Workflow · Connectors · Pipelines · Personal projects · Generative tools

Systems

  1. The dashboard's Open work tab on sample data: 44 items where the company owes the next move, 6 in the design queue, and 1 client reply to answer first.
    01 / Screen View

    Operations dashboard

    A company-wide dashboard, in daily use

    Brings tasks from three separate systems, plus open orders and who they belong to, into one view. Sign-in through Google Workspace, with what each person sees decided by their role.

    • Cloudflare Workers
    • TypeScript
    • Google Workspace sign-in
    View the code
  2. The dashboard's Briefing tab: a written briefing for the day that turns the order and task numbers into what to do first and who holds what.
    02 / Screen View

    AI daily briefings

    A scheduled writing job

    Claude writes a company-wide briefing and a personal day plan for each person from live task and order data, tiered so people read what is theirs rather than everything.

    • Claude
    • Anthropic API
    • Cloudflare Workers
    View the code
  3. The shared inbox: several team mailboxes in one list, unread client messages first, each tagged with the person whose inbox it came to.
    03 / Screen View

    Shared inbox

    One view over every team inbox

    Makes every inbox readable and answerable in one place for a short-staffed team, with sending gated by role and the right signature attached per inbox.

    • TypeScript
    • Cloudflare Workers
    View the code
  4. Diagram: Claude Desktop talks to the CRM server over stdio, and the server reaches Bitrix24 through a webhook, with 14 read tools, 16 gated write tools and the field schema read at run time.
    04 / Diagram View

    CRM connector

    A Model Context Protocol server

    Lets an AI assistant work with a CRM directly. It reads the CRM’s own schema at run time instead of hardcoding it, so it keeps working when fields and pipelines change underneath it, which they do.

    • Model Context Protocol
    • Bitrix24
    • Python
    View the code
  5. Diagram: Claude calls a read-only orders server, locally or on Cloudflare Workers, which paces its requests to the Plinth order API under one shared rate limit.
    05 / Diagram View

    Order backend connector

    A second Model Context Protocol server

    Covers the order system, which makes cross-system questions answerable: which of a rep’s leads became orders on a given day, for instance, without anyone exporting a spreadsheet.

    • Model Context Protocol
    • TypeScript
    View the code
  6. A terminal session: the weekly pipeline runs, then an audited email draft shows a confidence score of 88 with approval still empty, beside a draft that failed its audit.
    06 / Terminal View

    Outreach drafting

    A drafting system with a human in the loop

    Personalises client emails from public profile signals and scores how confident it is in each draft. Every draft goes through human review before it is sent, never straight out. Used across about 1,464 deals in three campaigns.

    • Claude
    • Anthropic API
    • Python
    View the code
  7. A terminal session: the pre-pass script sorts three client images by type, picks a Topaz model for each, and writes the prepped files, a report and a zip.
    07 / Terminal View

    Batch upscaling pipeline

    An image pipeline for print

    Takes low-resolution client artwork and AI-generated artwork up to large-format print quality in batches, with a prepass that prepares each file before the main model pass rather than feeding it in raw.

    • Python
    • Pillow
    • NumPy
    • Topaz
    View the code
  8. A terminal session: the pre-pass report for one transparent logo, recommending the Text and Shapes model with compression fixing, denoise and sharpening off, and saying why.
    08 / Terminal View

    Topaz settings skill

    A Claude skill

    Reads a client image and recommends which model and settings to use, working from Topaz’s own documentation rather than guesswork, so the choice is repeatable across people.

    • Claude
    • Topaz
    View the code
  9. An index of the twenty Claude skills, grouped as CRM and pipeline, files and print, web and catalog, and writing and setup, each with a one-line purpose.
    09 / Diagram View

    Workflow skill library

    Reusable skills for recurring work

    A set of skills for the jobs that come round every week: CRM routing and triage, outreach enrichment, production speed tables, mailer address formatting, artwork prep and SOP writing.

    • Claude
    View the code

Personal projects

  1. Cortex, a desktop AI client: a chat about a backup script, with the model, tokens in and out, cost and time under each reply, and the month's spend against a budget.
    10 / Screen View

    Cortex

    A desktop AI client

    A native desktop client for working with models directly, built to be fast to open and quiet to use.

    • Tauri
    • React
    • DeepInfra
    View the code
  2. Diagram of the transcription pipeline: a recording goes from an inbox folder through ffmpeg, local Whisper transcription, word alignment and speaker separation, and comes out as a transcript that says who said what.
    11 / Diagram View

    Transcription pipeline

    A local transcription pipeline

    Transcribes recordings and separates who is speaking, so a long multi-person recording comes out readable rather than as one undifferentiated block.

    • Whisper
    • pyannote.audio
    • Python
    View the code
  3. A terminal session: the security gate stages 21 notes from two sample vaults, stages nothing on an unchanged rerun, then stages only the one note that was edited.
    12 / Terminal View

    Knowledge memory pipeline

    A nightly distillation job

    Distills a large personal knowledge base into a queryable memory layer each night, passing everything through a security gate first so private material stays out of it.

    • Python
    • Obsidian
    • Notion
    View the code

Generative AI tools

Hands-on since 2023, on my own time.

Image
  • Midjourney
  • DALL-E
  • Stable Diffusion
  • Flux
  • Seedream
  • Gemini (Nano Banana)
  • Adobe Firefly
  • Grok
Video
  • Veo
  • Kling
  • Seedance
  • Wan
Design platform
  • Kittl