In reality, modern AI-powered products are usually composites: deterministic systems doing structured work, paired with large language models doing human-facing work. I built ConnectAI to make that distinction visible using a game everyone already understands.

Instead of writing another explainer post about decision trees vs. LLMs, I wanted a show-don’t-tell demo. So I used Connect 4!

You can play the app here:
👉 https://connect-ai-v1.web.app

And if you want a quick walkthrough first, here’s the short demo video I posted on LinkedIn:

The idea: make the architecture impossible to miss

ConnectAI is an educational web app that looks like a normal Connect 4 game at first glance. You drop pieces into a 7×6 board. The computer responds. You win, lose, or draw.

But right under the board, the app exposes what’s actually happening:

• An AI Engine labeled Decision Tree
• An LLM Chat panel labeled Gemini 2.0 Flash-Lite
• A “See prompt” link that reveals how the LLM call is constructed

That pairing is the lesson. A classic search-based AI doing structured reasoning, with an LLM layered on top for communication.

Three layers working together

The app naturally falls into three layers, each optimized for a different kind of intelligence.

1. Gameplay layer: the product surface

At the top is a clean, mobile-first Connect 4 experience:

• Standard 7×6 grid
• Turn-based play
• Immediate visual feedback as pieces drop
• Lightweight navigation (New, About, Author)

This layer is intentionally familiar. By removing cognitive overhead, the user can focus on how the AI behaves rather than learning new rules.

2. Decision layer: deterministic intelligence

When it’s the computer’s turn, the game state is handed to a deterministic decision engine.

Under the hood:

• Minimax-style adversarial search
• Alpha–beta pruning to cut off branches that can’t influence the final decision, improving performance
• A bounded search depth (8 plies) to balance strength and performance

The engine evaluates hundreds of thousands of possible future board states and selects a move assuming optimal opponent play. The UI even surfaces some metadata (depth searched, time taken, and scenarios evaluated) so users can see that this component is doing exhaustive computation rather than just guessing.

This approach is well-established in classical AI research. For two-player, turn-based, zero-sum games, deterministic search remains one of the most reliable techniques available.

Connect 4 is a particularly effective teaching tool here. Despite its simple rules, it’s a solved game—perfect play on a 7×6 board guarantees a win for the first player. That fact reinforces the core message: when a problem is well-defined and fully observable, structured algorithms are often the right tool.

3. Communication layer: language, not logic

Once the move is chosen, a second system takes over.

This is where the LLM comes in.

The app sends the current board state and the deterministic engine’s analysis to a server-side endpoint. A lightweight language model generates short, personality-driven commentary based on that context: trash talk, reactions, confidence, or frustration.

Crucially, the LLM is not deciding the move.

Instead, it’s doing what large language models excel at:

• Natural language generation
• Tone and style variation
• Human-like responses that would be tedious to hard-code

The “personality” slider modifies the prompt, and the “See prompt” link exposes that prompt directly. This turns prompting into a first-class engineering artifact rather than a hidden implementation detail.

That design choice aligns with modern prompt-engineering best practices: clear instructions, structured context, and explicit constraints dramatically improve LLM reliability and usefulness.

A modern hybrid AI system you can interact with

Zooming out, ConnectAI is really a miniature example of a modern hybrid AI architecture:

• Deterministic core
  Reliable, repeatable, optimized for structured reasoning.
• Generative interface
  Flexible, expressive, and human-facing.
• Explicit boundaries
  Each system is constrained to the job it’s best at.

By labeling these components in the UI and exposing the prompt, the app becomes an interactive system diagram. Users don’t have to trust an explanation, they can see the separation in action.

This mirrors a growing consensus in applied AI: the most effective systems combine classical algorithms with LLMs, rather than replacing one with the other.

Under the hood: the tech stack

ConnectAI is built on a modern, full-stack TypeScript foundation:

• Next.js for frontend and backend
• React for UI and state management
• Tailwind CSS + shadcn/ui for styling and accessible components
• Firebase Hosting for deployment and scalability

On the frontend:

• Game state is managed with useReducer for predictable transitions
• The board is an interactive SVG with animations
• The AI move computation runs in a Web Worker, preventing UI freezes during deep searches

The deterministic engine implements Minimax with alpha–beta pruning and returns metadata that’s surfaced directly in the UI.

On the backend:

• Next.js API routes handle LLM calls
• Structured game state and engine analysis are passed into the prompt
• The response is streamed back as real-time commentary

Firebase’s serverless hosting model keeps the deployment simple while allowing the app to scale without manual server management.

Where this started

This project actually traces back to coursework I completed during my Graduate Certificate in Applied Artificial Intelligence at UNC Charlotte. One of the early assignments involved building a Connect 4 AI using classical search techniques, which left a lasting impression on me.

That program does an excellent job of grounding modern AI discussions in fundamentals—search, optimization, and reasoning—before layering in machine learning and LLMs. I highly recommend it to anyone looking to deepen their understanding beyond surface-level tooling.

Program link here.

The bigger lesson

ConnectAI isn’t really about Connect 4.

It’s about demonstrating that:

• LLMs shouldn’t always replace deterministic systems
• Deterministic systems don’t need to be human-friendly
• The best AI products combine both, with clear, intentional boundaries

Instead of saying “don’t use LLMs for everything,” the app shows why.
The move engine is optimized and predictable.
The LLM is creative, bounded, and expressive.

That mental model scales far beyond games. And it’s increasingly critical as AI systems move from demos into real, high-stakes products.

Sources & Further Reading

1. Russell, S., & Norvig, P. Artificial Intelligence: A Modern Approach — Chapters on adversarial search and minimax
2. Knuth, D. E., & Moore, R. W. (1975). An analysis of alpha-beta pruning
3. Allis, L. V. (1988). A knowledge-based approach of Connect-Four
4. Google DeepMind. Prompt engineering best practices
5. OpenAI. Best practices for building LLM-powered applications
6. Browne et al. (2012). A Survey of Monte Carlo Tree Search Methods (contrast with deterministic search)

Manuel II of Portugal (1889–1932)

King Manuel II ascended to the throne of Portugal abruptly following the assassination of his father and older brother in 1908. His reign was short-lived, ending in 1910 when a republican revolution swept away the Portuguese monarchy. Manuel went into exile in England, settling near London. There, he lived a quiet, dignified life, largely withdrawing from politics and instead dedicating himself to cultural and charitable activities. Despite his exile, Manuel retained a love for his homeland and collected significant literary works, eventually building a substantial library reflecting Portugal’s rich cultural heritage (source).

Wilhelm II of Germany (1859–1941)

Wilhelm II, Germany’s last emperor, reigned until the defeat of Germany in World War I forced his abdication in 1918. Fleeing to the neutral Netherlands, Wilhelm lived in exile at Huis Doorn. In this quiet countryside estate, the former Kaiser retreated from public life, focusing his energies on writing memoirs and overseeing his gardens. He remained controversial, often blamed for provoking the war, yet lived in relative comfort until his death in 1941, far from the power struggles of Europe (source).

Constantine II of Greece (1940–2023)

King Constantine II was forced into exile after a failed counter-coup against Greece’s military dictatorship in 1967. Initially settling in Rome, Constantine and his family later moved to London, maintaining strong connections to the British royal family. Constantine’s life abroad was largely defined by low-key philanthropy and occasional public appearances connected to his royal heritage. Despite periodic returns to Greece, he never reclaimed his throne, and died without regaining his former position, though he remained a respected figure in European aristocratic circles (source).

Umberto II of Italy (1904–1983)

Umberto II, Italy’s last king, had a reign lasting only 34 days in 1946 before a national referendum abolished the monarchy. He left Italy permanently, settling in Cascais, Portugal, where he adopted a quiet and politically neutral lifestyle. He refrained from public political statements, dedicating his life to personal interests and supporting charitable causes. Despite living in exile for decades, he remained deeply attached to Italy, even if from afar, and was remembered affectionately by some Italians as “the May King” (source).

Simeon II of Bulgaria (born 1937)

One of the most remarkable examples of a monarch’s life after losing the throne is Simeon II of Bulgaria. Deposed by a communist regime in 1946 at just nine years old, Simeon fled to Spain, where he lived quietly for decades, becoming a successful businessman. In a dramatic return to his homeland after the fall of communism, Simeon re-entered public life, becoming Prime Minister of Bulgaria in 2001. His political comeback is unprecedented among modern European monarchs, representing an extraordinary transition from royal exile to democratic leader. This remarkable second act makes Simeon II a unique historical figure, highlighting how life after the throne can lead to surprising new beginnings (source).

LLM (Large Language Model)

An LLM is an AI that’s read and learned vast amounts of text and can predict what words come next in a sentence. Think of it as the world’s most advanced autocomplete tool—like the predictive text feature on your phone but infinitely smarter.

Neural Network

This is the technology behind LLMs, a network of artificial “neurons” (inspired by the human brain). Each neuron performs simple calculations, and together, they handle complex tasks, like the individual cogs in an incredibly complex machine.

Parameters

Parameters are essentially the adjustable knobs or switches within a neural network that the model tunes during training to learn from data. The more parameters, the smarter and more capable the model typically is.

Temperature

Temperature in LLMs controls how creative or predictable the model’s responses will be. A low temperature (close to 0) makes the AI more conservative and precise, while a higher temperature (closer to 1 or more) produces more diverse, imaginative responses.

Prompt

A prompt is simply the input you provide to an AI to guide its response. It’s like asking a clear question or giving specific instructions to someone to get exactly what you need.
When you use AI tools, the tool will likely add some additional context onto your prompt to guide the output. For example, when using an image generation tool, the tool will likely add something like, “Respond with a .png image file with 1024x1024px resolution” to each of your prompts.

Tokens

Tokens are chunks of words or characters processed by an AI. They’re like individual puzzle pieces the model assembles to form coherent sentences and paragraphs. Longer prompts or responses require more tokens.

Fine-tuning

Fine-tuning means taking a general-purpose AI and teaching it further on specific topics or tasks with specialized data. Imagine hiring an employee with general skills and training them specifically for your business’s unique needs.

Hallucination

Hallucination occurs when an AI confidently gives incorrect or nonsensical information. Imagine asking a confident colleague for directions, only to later realize they had no idea and just guessed convincingly. Remember, AI doesn’t know if it’s right or wrong!

Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard developed by Anthropic to enable seamless integration between AI systems, particularly large language models (LLMs), and external data sources and tools. By providing a universal protocol, MCP allows AI applications to access and interact with diverse datasets and functionalities without the need for custom integrations for each source. ​

Reinforcement Learning with Human Feedback (RLHF)

RLHF means humans guide the AI’s learning by rating its responses, teaching the AI to better match human expectations. It’s like training a new employee by giving regular feedback on their work to improve over time.

Context Window

The context window is how much recent conversation or instruction the model can “remember” when generating responses. It’s similar to your working memory—there’s only so much you can keep track of at once before you have to reference notes or past conversations. Having a very large context window can help generate more useful responses at the cost of more more processing power.

According to ESPN.com data as of February 23rd, 2025, Purdue leads all teams with 67.35% of its minutes played by in-state players. As a proud Purdue alum, it’s exciting to see the program excel in developing and utilizing local talent.

Top Teams by Percent In-State Minutes

Following Purdue are South Carolina (66.05%), Villanova (63.60%), and Minnesota (55.56%). Programs like Texas (54.85%) and Notre Dame (54.13%) also show strong reliance on homegrown talent!

Broader Landscape: Major Programs Breakdown

On the other end of the spectrum, some notable programs, including UConn, Missouri, Nevada, New Mexico, and West Virginia, recorded 0% in-state minutes.

Full Data Table: In-State Minutes Across All Teams

Final Thoughts

While this metric may not determine championships, it offers insight into recruiting strategies and local player development. For Purdue, leading in this category underscores the strength of basketball talent within Indiana. Go Boilers!

I’ve recently launched Content Machine, a free tool designed to streamline the content creation process. Whether you’re looking for fresh content ideas or want AI-generated blog posts and social media summaries, this tool makes it quick and easy.

Why I Built Content Machine

As the designer and creator of this tool, my goal was to demonstrate how OpenAI’s API can be leveraged to build practical, useful applications. Content Machine is an experimental project, free to use, with $20 of API credits preloaded so that users can try it out. If usage grows, I may explore adding ads to help cover costs.

If you’re interested in setting up a version with your own API key, have ideas for similar tools, or want to suggest additional features, feel free to reach out: palmerincarmel@gmail.com.

How It Works

Using Content Machine is simple:

1. Enter basic information about your content needs.
2. Choose from five different post formats.
3. Generate a full blog post instantly.
4. (Optional) Generate a punchy LinkedIn summary.

For convenience, users can also import/export their inputs. Clicking “Export” in the top right will save inputs to a text file. Later, clicking “Import” allows users to reload their saved inputs without retyping them.

Tech Stack & Workflow

Content Machine is built with a straightforward yet effective architecture:

• Front-End: Developed using HTML, JavaScript, and Firebase Hosting for a lightweight and responsive experience.
• User Input Handling: Users provide input, which is structured into a predefined prompt format.
• AI Processing: The formatted prompt is sent to OpenAI’s GPT-4o-mini model via an API request.
• Content Generation: OpenAI processes the request and returns the generated content.
• Output Display: The generated content (full blog post and optional LinkedIn summary) is displayed back to the user on the front end.

This seamless flow enables fast and efficient content generation while keeping the tool simple and accessible.

Creative Prompting Ideas

To get the most out of Content Machine, try experimenting with different prompt styles:

• Make the LinkedIn post a rhyming poem for a fun and engaging twist.
• Add lots of emojis to make your blog post or LinkedIn summary more expressive and eye-catching.
• Bold and italicize important topics to emphasize key points.
• Use simple language or tailor the writing style to match your target audience.
• Ask for a wide variety of blog ideas to explore different angles and creative directions.

Important Notes

• Experimentation Only – Content Machine is not intended for commercial use. Use it at your own risk.
• Data Handling – Your inputs are processed via OpenAI’s API.
• No Web Scraping – This version does not scrape the web, but I’m considering it for future iterations.
• Model Used – The tool currently uses GPT-4o-mini, which is fast and cost-effective, but not the most advanced model. If you’d like a version running on a different model, let me know.

Try It Now

Give Content Machine a try and see how AI can enhance your content workflow. I’d love to hear your feedback!

Doing More with Less

DeepSeek’s V3 model claims to deliver performance similar to ChatGPT-4o while using only $5 million in hardware and electricity, far less than the rumored $100 million spent by some competitors. The energy demand for AI is so high that some companies are even considering nuclear reactors to power their data centers.

Mixture of Experts (MoE)

The key innovation behind DeepSeek’s efficiency is its use of Mixture of Experts (MoE). Unlike a single massive model handling all tasks, MoE uses specialized smaller models, making it far more efficient while maintaining high performance.

Chain of Thought for Complex Reasoning

DeepSeek’s R1 model employs Chain of Thought (CoT) reasoning, helping it tackle multi-step problems more effectively. This is similar to OpenAI’s o1 model, enabling better logical processing compared to base AI models.

Open Source Advantage

Unlike OpenAI’s models, DeepSeek is open source, meaning anyone can download and experiment with it. This transparency allows developers to explore its capabilities freely.

The Caveats

Despite its promising claims, there’s skepticism about DeepSeek’s actual efficiency and performance. Still, the results have been promising enough to deflate the US tech market. Also worth keeping in mind, as a Chinese company, DeepSeek operates under government censorship, which may limit its access to certain information and influence its outputs.

Final Thoughts

DeepSeek’s advancements in AI efficiency and openness are exciting, but questions remain about its true performance and potential limitations. What do you think?

Sources

• Computerphile’s YouTube video
• Hugging Face Download DeepSeek V3
• Reuters - Microsoft, Meta back big AI spending despite DeepSeek’s low costs
• CNBC - Why tech giants such as Microsoft, Amazon, Google and Meta are betting big on nuclear power

This accomplishment stands out as other discount airlines, like Spirit Airlines and JetBlue, face significant challenges. As a frequent traveler from Raleigh-Durham International Airport (RDU), my personal experiences with Breeze have been overwhelmingly positive, and I believe their success is in large part down to three critical strategies.

1. Scaled Sensibly

Breeze took a smart approach to growth by leasing smaller Embraer 190/195 planes during their early days. This allowed them to expand to underserved markets quickly while waiting for deliveries of their game-changing Airbus A220-300s. These A220s stand out for their fuel efficiency, passenger comfort, and operating cost advantages, helping Breeze secure a sustainable edge in the competitive airline industry.

2. Leveraged Their Network

David Neeleman, Breeze’s founder and CEO, tapped into his deep aviation expertise and relationships to build a robust operation. For instance, leasing planes from Azul Brazilian Airlines, which Neeleman also founded, was a key strategic move that helped to bridge a capacity gap. Moreover, Neeleman recruited key leaders from successful competitors, especially Allegiant, which has profitally operated using a similar point-to-point network strategy.

3. Kept It Simple

From my perspective, simplicity is where Breeze truly excels. Their website and app are intuitive, and their fare bundles are transparent. This eliminates the surprise fees or confusion at the check-in counter that often accompany discount airlines, leading to a calmer, more enjoyable travel experience. Dare I say it’s a breeze?

Most flight packages include free Wi-Fi, setting Breeze apart in the budget airline sector. Additionally, their rewards program is straightforward and easy to use, with no complex tiers or blackout dates. This commitment to simplicity means passengers board relaxed, and it reflects in the overall vibe of the flights I’ve taken from RDU.

Looking Ahead

Breeze Airways has set a strong example of how thoughtful strategy and customer-focused operations can lead to profitability, even in challenging times. As someone who flies Breeze regularly from RDU, I am optimistic about their future and look forward to seeing them maintain their success while continuing to deliver exceptional travel experiences.

What has your experience been flying with Breeze?

Sources

• The Points Guy - Breeze Airways’ Profitable Quarter
• Breeze Airways Press Release - First Profitable Quarter
• Breeze Airways - Leadership Team
• FlightGlobal - Breeze Profit Announcement
• Airline Weekly - The U.S. Airlines Growing and Shrinking in 2025

Furthest Range for a Single Aisle Jet

The main distinguishing characteristic of the A321XLR is the 4,700nm range. That’s 27% further that the popular existing A321neo model introduced into service in 2017. This gives the A321XLR the furthest range of any single aisle commercial jet.

Impact on Midsize Airports

For midsize airports like RDU, having a medium size aircraft capable of flying across the Atlantic could be a game-changer. Currently, RDU has daily transatlantic service to London, Paris, and Frankfurt on wide-body planes. These are popular destinations that are also major hubs for connecting passengers to other parts of Europe and Asia.
But, the lower capacity A321XLR could make it financially viable for airlines to serve “skinnier” routes to destinations like Barcelona, Munich, or Rome. It could even be used to service South American destinations like São Paulo.

Development and Rollout

The A321XLR was first announced in 2019 with the first delivery being made in 2024 to Spanish carrier Iberia Airlines. In the US, both American and United have orders for 50 of the jets, with JetBlue having an additional 13. While the airlines have indicated that they’ll likely start the planes on longer domestic routes, there’s no telling what kind of unique international routes could be opened up in the future!
Technologically, there’s a few key advancements that made this extended range possible. First and foremost, an additional fuel tank is installed in the rear of the aircraft, providing additional fuel capacity. Most airlines will opt for a premium-heavy seating configuration, thereby reducing overall seating and extending range further.
To accommodate the extra fuel weight during takeoff, the plane has a reinforced landing gear, as well as a aerodynamic improvements like optimized wing flaps.

Conclusion

Overall, the A321XLR represents an exciting opportunity for small and medium airports like RDU. What are your thoughts on the A321XLR? Are there any routes from your home airport that would be good candidates for this plane?

Sources

• Airbus
• Wikipedia
• Great Circle Mapper
• ThePointsGuy
• Aviationa2z

Parking and Transportation

Unlike most larger airports, TXK’s parking, passenger drop off, and car rentals are conveniently just steps from the terminal. Don’t miss the water feature!

Passenger Drop Off

I doubt you’ll ever hear honking or need to circle the building while waiting for passengers to arrive here!

Parking

If you aren’t lucky enough to have a friend or family member drop you off, fear not! TXK has covered and uncovered parking available starting at $10 per day.

Car Rentals

TXK is served by Budget, Avis, and Dollar car rentals. No need to worry about catching a shuttle, the rental lot is adjacent to parking.

Ground-Side Terminal

Whether departing or arriving, passengers will surely enjoy the clean, tastefully designed terminal building. Combined with the absence of garish advertisements and friendly service at airline/car rental check in counters, the atmosphere is comfortable, quiet, and relaxed. Security opens up about 90 minutes before scheduled flights, and took me about 10 minutes the day I was there. Make sure not to cut anyone in line, there’s a good chance you’ll be seated by them on your flight!

Air-Side Terminal

Once through security, a short escalator ride takes you up to the departure gates. I would wager this is one of the only escalators in Texarkana! There is also an elevator available. Upon reaching the top, you’re greeted by a dramatic illustration of one of Texarkana’s most interesting features. If you hadn’t guessed by the name, Texarkana lies squarely on the Texas-Arkansas border. The local post office (pictured) is split down the middle by the border… sounds like a beuracratic nightmare!

TXK’s two gates share an appropriately-sized waiting area equipped with reclined seating, free WiFi, and plenty of power outlets.

Additionally, around the corner there’s a more casual lounge area with well-stocked vending machines. Floor-to-ceiling windows spanning the entire gate area give an expansive view of the airport’s primary runway and FBO across the way. I watched as the American Airlines CRJ-900 I would be taking to DFW landed and quickly taxied to the gate. As we pushed back from the gate, our pilot sarcastically quipped that we were first in like for takeoff!

Destinations

Currently, TXK is only served by twice-daily service to DFW operated by American Airlines affiliate American Eagle. As recently as 2022 United maintained service to Houston, hopefully that will be resumed in the near future.

Conclusion

If you find yourself visiting the TX/AR/LA/OK quad state area consider flying into Texarkana Regional! While TXK may not have the glitz and glam of larger airports, you’ll certainly be charmed by the friendly service and convenience of this small airport.

Visit TXK’s website to learn more about the airport, and take a look at Texarkana’s Wikipedia page to learn more about this unique city, including the rivalry between the Texas High Tigers and the original Razorbacks of Arkansas High.

With conference realignment in full swing, how do new Big Ten and SEC schools fit in with their peers in terms of total enrollment and endowment? Check out the chart below to see! (note that the y axis is in logarithmic scale)

Sources

Big Ten SEC