From Map to Playbook: My AI Journey Behind Mach Maglev

---

The spark

In college, I took an “AI in business” class, where we were tasked with creating a fantasy

company and use our Artificial Intelligence (AI) knowledge that we gain from the semester to implement tools into the company. For me, I wanted a fast way to

move me and my stuff from my home in New York to college in Florida, taking inspiration from Amtrak’s Auto Train, I thought of a futuristic maglev service to transport people and cargo long distances.

I didn’t start with a finished plan, I started with a corridor and a question. If AI helps us plan and operate rail differently, what would a mixed passenger/freight line actually look like? The Ontario‑Atlantic Super Train (aka Mach Maglev) became my fantasy startup: a safe sandbox to design, test, and show my work across the full stack: strategy, operations, marketing, finance, ethics, and customer service.

The corridor decision

The route (St. Catharines → Buffalo → Rochester → Jacksonville) forced real trade‑offs. Mixed passenger/freight means balancing speed and load, commuter needs and freight demand, layover times, and cross‑border friction. In the early overview materials, I framed the corridor as connecting strategic tech hubs in Ontario and Western New York to a deep‑water port ecosystem in Florida, then captured the obvious tension: ambition vs. cost and complexity.Mach Maglev is designed for zero emissions. As a concept, that phrasing matters. It sets an environmental goal without over‑promising the parts that depend on future energy mix, regulatory approvals, and procurement.

The AI stack I built

I built out three use cases first because they read clearly to operators, executives, and passengers:

• Routing optimization. AI helps propose timetables and freight slots that reduce congestion while keeping passenger service reliable. The idea wasn’t to ‘let a model run the railway,’ but to surface candidate schedules and let planners make the call.

• Predictive maintenance for the guideway. I drafted a loop: IoT sensors detect track defects and AI flags sections for inspection; the system auto‑drafts an FRA‑style report for crews; maintenance runs continuously rather than as a stop‑start scramble. The point is simple—shorter detection‑to‑fix time and fewer surprises.

• Customer‑facing AI agent (Speedy). I built a Q&A template so an agent can answer the obvious public questions quickly: what Mach Maglev is, what regions it serves, why it’s faster, and how AI supports routing and maintenance. An agent isn’t a press office; it’s a front door that gets people oriented before they talk to a human.

Across all three, human oversight stayed in the loop. I reviewed and edited outputs, especially in marketing and AI video generation, to correct tone, visuals, and any claims that drifted beyond the concept.

Marketing Goals

I didn’t use AI to “do marketing.” I used it to think through marketing faster. The five‑week frame gave me a clock and a sandbox; AI helped me find an audience, shape the message, and preview what a campaign might do if it were real. Then the human part began: cutting copy, tightening claims, and deciding what stays. The goal was to learn how the system behaves when you ask it good questions.

The first lesson was prompt specificity. Vague inputs produced vague campaigns. When I told the model who we were talking to: logistics executives, infrastructure investors, policymakers, and what they actually cared about (congestion, fuel costs, cross‑border friction), the outputs moved from generic hype to something a buyer

could recognize. AI drafted options quickly, but I had to rewrite for clarity and cut any phrasing that sounded like a slogan. That’s how the tone settled into “confident, practical, future‑focused” instead of “AI‑sounding.”

The second lesson was iteration beats cleverness. I used AI to generate variations such as subject lines, hero lines, and short scripts. These were measured them against simple guardrails: Would I say this in a boardroom? Could I defend this claim in an Q&A? If the answer was no, it didn’t ship. Those guardrails kept the numbers honest, too. Metrics such as conversion and engagement targets stayed in the plan, but as yardsticks, not trophies. They helped me compare options, not declare victory. Every time a draft edged into over‑promising, I pulled it back to the phrase that keeps the concept credible: designed for zero emissions.

The third lesson was that Human-AI-Human interactions matters most when stakes are reputational. AI can score leads and power a front‑door agent, but it can’t decide which message represents the company. I kept human review at the beginning and end of each cycle, especially for video and social media campaigns. Checking tone, claims, and visual accuracy before anything “went live” in the portfolio is essential for real life operations. That rhythm, AI drafts, I decide, is what made the campaign artifacts usable rather than noisy.

Finally, I learned that bias and trust travel with the message. Even in a fantasy startup, you must think like an operator. If personalization skews who sees which claims, or if copy ignores equity, you’re planting future risk. My approach was straightforward: keep the message inclusive, avoid sensational claims, and make the ethics posture part of the story (data handling, environmental constraints, and human checks where stakes are high). In practice, that meant editing out loaded language, verifying any automation claims against the concept’s limits, and staying transparent about what’s simulated.

What did AI actually change? Speed and range. It widened the exploration space: more angles, more lines, more micro‑pitches. AI also gave me enough signal (goals, KPIs, timelines) to choose responsibly. What remained human was everything that defines the brand: judgment, tone, and the choice to say less when more would only sound good. The campaign is part of a learning playbook, not a scoreboard, and that’s exactly what I needed.

Operations blueprint: the maintenance loop

AI can help in operations and project management from a high level standpoint. The Process Optimization Plan breaks the maintenance problem into practical pieces:

• Pain points today: red tape, little parallelization, long detection‑to‑fix windows, manual reporting.

• AI opportunities: predictive flags for sections that need inspection; IoT sensors on trains to catch defects in motion; auto‑drafted reports to cut paperwork friction.

• Implementation path: assess current workflows, train crews, pick a pilot site, run the pilot, expand after what’s proven.

Risk & mitigation: report quality (co‑develop with the regulator), misclassification (diverse training + manual audits), sensor failure (backups + calibration).

• KPIs: detection‑to‑repair time, detection accuracy, reporting time, cost savings, crew productivity.

It’s not a fancy plan, it is a loop that turns maintenance into a continuous process instead of a costly emergency.

Financial discipline

AI generated a simple financial forecast with conservative monthly figures:• Revenue: ~$200K–$260K per month across the visible forecast period.• Expenses: decline after February due to loans paid off, then trend in the $185K–$230K range depending on month.It’s not about proving profitability on paper; it’s about showing how assumptions affect runway and risk tolerance. Keeping numbers conservative avoids the trap of building a narrative on optimistic math you can’t defend later.

The Importance of Ethics

I focused on three kinds of bias because they land directly on risk, trust, and reputation:

• Hiring bias in HR automations. Automated screening can skew based on historical data. Don’t let a model quietly shape your workforce; keep humans in the loop and audit outcomes for gender and race bias.

• Border/security caution at high stakes. Use AI for classification and triage, but keep human verification for customs and national security decisions. Don’t over‑delegate where the cost of a mistake is high.

• Data security: staff training to avoid leaking sensitive info; strict guidance on what goes into models.

• Environment: be mindful of the energy and water footprint of AI infrastructure; the phrase ‘designed for zero emissions’ keeps the focus on design intent while you build out credible energy plans.

Emphasis was focused on the data given to AI. If AI systems are given biased data, then the model is going to have bias. Unfortunately, whether it is intentional or not, a lot of data given to AI has bias. The most important step for any AI implementation is to ensure the data received by your model is bias free. If your AI looks at people, give the model people of every shape, size and color. If the model needs example of a “good” resume, ensure the data doesn’t accidentally attribute gender-based traits as bad. The list goes on. The most important key to AI is the data received, and we have a responsibility to try our best to keep that key clean.

What I learned

• AI video & Image generation. It’s fast, but raw. The first cut can drift in tone or claim more than the concept allows. Reviewing, rewriting lines, and swapping visuals made the difference between a slick clip and a credible one. Even trying to generate a cover image for this blog is in this category, careful prompt engineering with some human tweaking afterwards makes a large difference to separate your AI content from the “AI Slop”

• Future analysis. I stopped trying to predict the future with confidence and started sketching scenarios. A corridor like this doesn’t hinge on one perfect forecast; it depends on how well you plan for ranges and choose the next reversible step.

In both, the pattern repeats: AI drafts, I decide. That’s the posture that kept this whole project grounded.

Want to learn more?

If you want to see the artifacts mentioned and watch AI in action, browse the AI page where every artifact lives: https://www.frankvisca.com/learn-more-ai-work-here

Learn more here!