High Value Project Tutorials - Cost Reduction Information

Cost Reduction Factors

Cost Reduction Factor Without DataPredict With DataPredict
Expertise Requires Hiring Of A Machine Learning Engineer, Starting At 5K USD Per Month To Build Optimized Models Required By Your Projects. Preferrably Holding A PhD Or Masters Degree To Ensure Correctness. Can Use Existing Game Developers. Can Learn From DataPredict’s High-Value Project Tutorials To Specialize Further.
Integration Additional Costs From Hosting External Servers And Connecting ML Frameworks In Different Programming Languages. Lua Only.
Compute Cost If Models Are Hosted Externally, Compute Cost Increases As Number Of Players Increases. Utilize Roblox Servers’ And Players’ CPUs At No Additional Cost.
Choosing Models Need To Learn Multiple ML Frameworks For Machine Learning, Deep Learning And Reinforcement Learning In Different Programming Languages If Models Are Hosted Externally. Unified Under Same API Design. All Models With High-Value Use Cases Are Covered.
Time-To-Market Requires Multi-Year Engineering For The Whole Library If Building From Scratch. Minimum Of 1 Week If Using Existing ML Frameworks Hosted In Different Programming Languages. Within A Day.
Latency Network Latency And Serialization Issues If Hosted Externally. Latency Only Comes From Models’ Training And Predictions. Mostly Depends On Roblox’s Servers And Clients.
Security If Hosted Externally, Vulnerable To Hacking. Data Stays Between The Roblox’s Servers And Clients.
Reliability May Be Inaccurate Or Underperform When Building From Scratch. If Hosted Externally, Models May Be Impacted By Third Party Servers. 3-Years Of Development.
Legal If Hosted Externally, Requires Active Compliance For Data Protection Laws. Data Stays Between The Servers And Clients. As A Result, Mostly Complies With Data Protection laws.
Domain Specialization Machine Learning Engineers Generally Do Not Specialize In Games. Built With Game Domain Specialties And Optimizations.

Licensing Costs

License Type Based On Gross Revenue Per Company / Organization / Group / Individual Percentage To Pay Based On Project’s Gross Revenue In USD Additional Notes Licensing Cost When A Project Earns 5K / 10K / 100K / 1M USD Per Month
Non-B2B + Less Than Or Equal To 5K USD Within 365 Days (Not Per 365 Days) 0% Requires Public Disclosure Of DataPredict Not Applicable
Standard 2% That Decreases As Gross Revenue Increases Requires Public Disclosure Of DataPredict 89 USD / 160 USD / 1300 USD / 10,530 USD
White-Label 5% None 250 USD / 500 USD / 5000 USD / 50,000 USD
  • Note: At the $5,000/year revenue mark, the licensing fee is approximately $8/month for “Standard” license. Collecting micro-payments of this size creates administrative overhead that exceeds the revenue itself. Therefore, we subsidize 100% of costs for organizations under this threshold. We only begin billing when your success generates a fee large enough to be meaningful for both parties.

Development Costs

The table assumes that only one person is working on each library at machine learning engineer average salary rate in the United States.

Library Personal Time Cost Work Hours Time Cost (40 Hours Per Week) Personal Monetary Cost (USD) Work Hours Monetary Cost (USD)
DataPredict™ 25000+ Hours (Spread Over 3 Years) 625 Weeks (~12 Years) ~300,000 ~1,260,000
DataPredict™ Neural 17000+ Hours (Spread Over 2 Years) 425 Weeks (~8.2 Years) ~200,000 ~840,000
DataPredict™ Axon 4000+ Hours (Spread Over 6 Months) 100 Weeks (~2 Years) ~50,000 ~210,000
DataPredict™ Zero 2000+ Hours (Spread Over 3 Months) 50 Weeks (~1 Year) ~25,000 ~105,000
All 48000+ Hours (Spread Over 5 Years) 1095 Weeks (~21 years) ~575,000 ~2,204,000

This assumes that the machine learning engineer work at these timing:

  • Weekdays: 7 pm to 3 am. Roughly 8 hours of work.

  • Weekends: 9 am to 3 am. Roughly 18 hours of work.

  • Weekly Total: Roughly 76 hours of work.

Factors That Led To These Pricing Strategies

  • DataPredict was initially built from passion as a fresh graduate, but eventually went to commercial licensing to allow it to be capable for production-grade use. Hence, the first year of development costs are considered “subsidized” due to it being more of a learning experience more than building a serious project.

  • I am based from Malaysia, and I can keep the costs low compared to the United States.

  • Being a solo developer means that there is no overhead from managing people, keeping the costs low.

  • I rely on the community to search for potential bugs in addition of my own bug hunting, keeping the costs low by not hiring a dedicated tester.

To understand why I don’t use AIs to let the AI to do full code generation to further lower the costs, you can have a look here.

Comparison To The Solution Of Other Game Engines

ML For Games Solution Description Features First Development Date First Release Date Development Duration For Production Release
Unity’s ML-Agents Uses Unity SDK, PyTorch And External Communicator To Train AI Agents. PPO, SAC And Imitation Learning. 19 September 2017 1 May 2020 955 days (2.7 Years)
Unreal Engine’s Learning Agents Uses Unreal Engine’s Blueprint System And C++ library To Train AI Agents. PPO And Behavior Cloning. September 2023 Still At Experimental Phase N/A
Godot’s RL Agents (Unofficial) A Wrapper For StableBaselines3, Sample Factory, Ray RLLib And CleanRL To Train AI Agents. A Variety Of Reinforcement Learning Models. 18 October 2021 Stalled At 26 February 2025 N/A

This site uses Just the Docs, a documentation theme for Jekyll.