Mark Salisbury
AI Prompt Engineer & Workflow Automation Specialist
I build AI tools that solve real problems for real businesses. Over the past year, I've designed and deployed custom AI solutions for small businesses in construction, service industries, and operations — saving hundreds of hours and generating measurable revenue.
Get In Touch
Case Studies
I don't just write prompts. I architect systems, iterate based on user feedback, manage platform risk, and train teams to use AI effectively. My work spans ChatGPT, Claude, Gemini, and custom automation via Google Apps Script with API integrations.
Current Role: Operations Manager | Company: StarForge Dynamics (Founder)
Location: Killingly, CT | Contact: [email protected]
1. Personal BrandForge: Voice Synthesis from Meeting Transcripts
The Problem
Every time AI generated content for me, I spent 15+ minutes rewriting it to sound like me instead of generic AI. I was tired of the "AI voice" — overly formal, weirdly enthusiastic, structurally predictable. I needed AI that could write in my actual voice.
The Solution
I built a system that analyzes my natural communication patterns and synthesizes them into AI-generated content:
  • Fed ~18,500 words into a Claude Project: meeting transcripts, emails, text messages
  • Identified the gap: my speaking voice ≠ my writing voice
  • Built a hybrid approach: pulled specific examples from transcripts + created guidelines for situational context (team communication vs. customer outreach)
  • Validated with my wife of 20 years — she couldn't tell the difference
  • Ported the system to ChatGPT custom GPT for comparison testing

Technical Approach:
  • Multi-source corpus analysis (transcripts, emails, texts)
  • Context-aware voice synthesis (adapts tone based on audience)
  • Cross-platform validation (Claude Project vs. ChatGPT custom GPT)
The Result
2
Build time
hours
95%
Rewrites eliminated
used to take 15 min/piece, now takes 0-2 min
Use case expansion: Now using it for all customer outreach, team communication, and marketing copy
This tool became the foundation for BrandForge, the voice engine that powers PostPulse (see below).
2. PostPulse: Automated Social Media That Sounds Human
The Problem
Small business owners don't have time to post on social media consistently. Hiring a social media manager is expensive. Generic AI content sounds like generic AI content. They needed automation that maintained their authentic voice.
The Solution
I built PostPulse, an automated social media system that generates and publishes content across Facebook, Instagram, LinkedIn, and X (Twitter) while preserving the client's brand voice.
How it evolved:
01
Phase 1 (Personal tool)
Custom GPT that generated social media calendars for my own business
02
Phase 2 (Feature expansion)
Added Ideogram-tuned image prompts, video scripts, thumbnail generation
03
Phase 3 (Content repurposing)
Built an engine to take sales call transcripts or podcast appearances and repurpose them into social content
04
Phase 4 (Production)
Partnered with developer Dan to build OAuth infrastructure and multi-platform posting automation
My Role:
  • Designed the AI/prompt layer (all content generation, voice synthesis, repurposing logic)
  • Created BrandForge: the voice engine that captures each client's communication style
  • Owned product strategy, naming, and marketing
  • Collaborated with backend developer for production infrastructure

Technical Approach:
  • Custom GPT → Gemini gem → production API integration
  • BrandForge voice analysis (similar to Personal BrandForge case study)
  • Multi-platform content adaptation (each platform has different best practices)
  • Google Apps Script automation for posting workflows
The Result
2
Customers
active, 3rd onboarding
2-10
Time saved
hours/week per customer (depending on posting frequency)
4
Platforms
Facebook, Instagram, LinkedIn, X
Revenue: Productized and selling as SaaS tool
PostPulse is now a core offering at StarForge Dynamics.
3. ApplyForge: Job Application Accelerator
The Problem
I applied to 1,500+ jobs over 9 months. Writing custom resumes and cover letters for every application was soul-crushing and time-intensive. I was spending 20 minutes per resume tweak and up to an hour creating custom applications. I needed speed without sacrificing quality.
The Solution
I built ApplyForge, a two-stage AI system that generates ATS-optimized, custom resumes and cover letters in seconds.
How it works:
Stage 1 (Job Vault)
User provides their complete work history. AI exhaustively lists every job and every duty they've ever performed, then stores it.
Stage 2 (Application Forge)
User pastes a job description. AI reads the description, matches relevant duties from the Job Vault, renames them with proper industry nomenclature (without lying or embellishing), and generates a custom resume + cover letter.
Platform iterations:
  • Started as a super prompt (copy/paste text)
  • Evolved into a ChatGPT custom GPT (one-click workflow)
  • Split into two Gemini gems (Gemini couldn't handle both stages in one tool)
  • Added ATS optimization based on deep research into applicant tracking systems
  • Built creative vs. technical mode (adapts resume format based on job type)

Technical Approach:
  • Multi-stage prompt architecture
  • Platform constraint adaptation (split tasks when single-model architecture failed)
  • ATS keyword optimization without keyword stuffing
  • User-friendly for non-technical job seekers (made it for friends who weren't AI-savvy)
The Result
1000+
Applications generated
1
Interviews before ApplyForge
(it was a scam)
9
Interviews after ApplyForge
Time per application: 20 min–1 hour → 0 minutes (just copy/paste job description)
I eventually landed my current role using ApplyForge. Now I've productized it for other job seekers.
4. ReddingtonGPT: AI Therapy Bot for Personal Growth
The Problem
I'm neurodivergent (ADHD + autism + RSD). One of my biggest blockers was asking for help. I would spin my wheels solo for months rather than reach out to my team or mentors. Traditional therapy helped, but I needed something more frequent, more accessible, and more direct.
The Solution
I built ReddingtonGPT, a custom GPT therapy bot with the persona of Raymond Reddington from The Blacklist — brilliant, strategic, emotionally intelligent, and willing to call me on my bullshit.
Strategic Reframing That Cuts Through Internal Narratives
When I was overthinking a team meeting, ReddingtonGPT reframed it:

"You are not simply managing a sales pipeline. You're orchestrating belief systems under pressure.
Here's a story. I once advised a Turkish smuggler who decided to go legitimate, open a chain of boutique olive oil stores—yes, really. First day, his sales team shows up in tailored suits, glistening smiles, ready to move product. He walks in, takes one look at them, and says: 'You're not selling oil. You're selling trust. If they don't believe you'd bathe your children in this oil, they won't buy it.' Then he fired half of them.
Trust, Mark. That's your currency tomorrow.
Not just the market's trust in StarForge—but your team's trust in you. That you have the map, even if it's drawn in blood and ink. That you see further, feel deeper, and will not let them drift into obscurity while you chase legacy."
This kind of strategic storytelling helped me reframe anxiety into action. It's why I built ReddingtonGPT
Why Reddington?
  • Direct communication style that cuts through my internal narratives
  • Strategic mindset that reframes problems as solvable challenges
  • Warm but firm — doesn't coddle, but doesn't shame
  • Fictional character, so I could be vulnerable without human judgment anxiety

Technical Approach:
  • Deep character analysis prompt (speech patterns, problem-solving approach, emotional intelligence)
  • Therapy framework integration (CBT, narrative therapy, Socratic questioning)
  • Context retention across sessions (uses ChatGPT memory features)
  • Built a second version based on real therapist Terrance Real for additional perspective
The Result
Time using it: 6 months
Breakthrough: Got over my fear of asking for help → reached out to my StarForge team → launched phase 2 of the company
Personal growth acceleration: Years of progress in months
This tool became essential to my daily workflow. I use it to process difficult decisions, challenge my assumptions, and stay accountable.
Note: I'm comfortable sharing this publicly because mental health stigma is bullshit, and AI-assisted therapy is going to be a massive part of accessible mental healthcare.
Skills & Tools
Core LLMs
ChatGPT (custom GPTs, API), Claude (Projects, API), Gemini (gems, deep research), DeepSeek, Qwen
Image/Video Generation & Audio
Ideogram, Gamma, Kling.ai, ElevenLabs. Descript, Napkin.ai, NotebookLM
Automation & Development
Google Apps Script (heavy production use), Airtable, Notion, API integrations
Specializations:
  • Multi-model validation (run queries across 4+ models for accuracy)
  • Platform risk management (learned the hard way when ChatGPT updates broke my tools)
  • Voice synthesis & brand personality engineering
  • Workflow optimization & chain-of-custody analysis
  • ATS optimization & job market strategy
Domain Expertise:
Construction operations, small business workflow automation, customer service systems, advertising/copywriting
What I Bring to the Table
Production experience
I've built tools for paying customers, not just demos
Platform fluency
I know when to use ChatGPT vs. Claude vs. Gemini, and when to split tasks across models
Systems thinking
I see workflows as chains and optimize weak links
Cross-functional collaboration
I work with backend developers, designers, and non-technical users
Relentless documentation
Every prompt iteration is tracked and versioned
User-focused iteration
I build, test, refine based on real feedback
I don't just write prompts.
I architect AI solutions that work with human workflows, not against them.
Let's Work Together
I'm open to full-time Prompt Engineer roles, contract work, or freelance projects.
Location
Killingly, CT (open to remote work)