If you're a developer juggling a 9–6 and side projects, you probably know this graph story too.
In 2025, my GitHub contribution graph told the tale of two different developers. The first half of the year: Empty squares. I did my job, logged off, and disconnected. The second half of the year: 4,000+ commits.
What changed? I stopped worrying about the "grind" and started leveraging AI to accelerate my build speed.
The "MVP Problem" is Solved
I remember trying to build an MMORPG side project years ago. The sheer amount of boilerplate and setup required meant I had to sink hundreds of hours before I even had a "Minimum Viable Product." It killed my motivation before I could even start.
With the current state of AI coding tools, that friction is gone. I discovered that I can now spin up an MVP in days, not months.
This new velocity gave me the momentum to:
- Deep dive into Backend Node.js and MongoDB.
- Master Nginx and VDS configuration from scratch.
- Actually ship a complete, production-ready project.
What I Built: An AI Companion Platform
I used this new velocity to build a platform centered around a core idea: AI Chatbots that feel alive.
I didn't want a generic assistant. I wanted to create conversational partners—friendly, romantic, smart, or supportive. To achieve this "human-like" feel, I engineered features that go beyond simple text generation:
Persistent Memory: The AI remembers context, past conversations, and small details about you to maintain a continuous relationship.
Rich Media: The bots can send photographs.
Interactive Gameplay: I didn't stop at text. I developed 4 mini-games (hosted on separate pages) where you can play and interact with the AI personality.
True Personalities: Each AI has its own backstory, hobbies, and character traits. They aren't just LLM wrappers; they are unique identities.
Why DeepSeek?
For the tech stack, I had to choose the right LLM. After testing various models, I settled on DeepSeek as the core intelligence for these characters.
I found that DeepSeek handled roleplay, emotional nuance, and "human" distinctiveness significantly better than the alternatives for my specific use case.
(I am currently writing a detailed technical breakdown comparing DeepSeek vs. OpenAI vs. Gemini benchmarks, follow me if you want to see the results!)
The Result
It has been a year of massive growth. 2025 was about learning to build fast. 2026 will be about refining that process and bringing this energy to my main work.
You can see the live project here: rizae.com
What's your favorite AI tool for side projects? Share below 👇
Happy coding!
Top comments (1)
DeepSeek + Node is a smart combo for an AI SaaS specifically because of the economics - DeepSeek's pricing means you can run inference-heavy features without the per-call cost murdering your margins, which is usually the thing that quietly kills AI SaaS businesses. A feature that's magical at 100 users and bankrupting at 100k is a real failure mode, and choosing a cost-efficient model up front is half the battle.
The scaling detail I'd watch as you grow: cost-per-active-user, not just total inference cost. The trap is a few power users whose usage is 50x the median, and a flat subscription that doesn't cover them. The fix is the same routing/caching discipline at the app layer - cheap model for the routine calls, cache repeated context, reserve the expensive path for what needs it. That's exactly the economics I obsess over in Moonshift (a multi-agent pipeline that ships a prompt to a deployed SaaS) - multi-model routing is what keeps a full build ~$3 flat. Great workflow writeup. As you scale, are you seeing cost stay linear with usage, or are power users skewing it? That curve shape decides your pricing model.