Learn how Partner Utopia transformed the partnership lifecycle with AI-driven innovation—and how Divelement helped them deliver a fully functional product under a tight 12-week deadline!
Partner Utopia provides a revolutionary partner management platform that automates and streamlines the strategy phase of the business partnership lifestyle. The platform uses proprietary AI (artificial intelligence) backed by the expertise of thousands of expert partner managers to assist clients with partner selection, partnership creation, and growth optimization.
When the Partner Utopia team had only 12 weeks to build their first fully functional product for an important industry conference, they turned to Divelement for help.
Partner Utopia’s founders knew they had a brilliant idea: a new tool that could automate the strategy and onboarding process for new partnerships. Although they had no tangible product yet, the company saw an opportunity to present the concept at an important industry conference. However, they faced one major problem — the conference was only 12 weeks away. Partner Utopia needed a fully functional MVP (minimum viable product) to launch their solution but lacked the internal resources to create one in the short time available.
The most obvious challenge was delivering a presentable, fully functional program in such a tight timeframe. Partner Utopia had about 12 weeks in total to train multiple AI models and build, test, and deploy the software platform.
Another major challenge was a lack of usable data to train and fine-tune the core AI models. Partner Utopia’s principals spent decades honing their partner management skills to become experts in the field, but they didn’t have a lot of machine-usable data for AI models to analyze. That meant they needed to generate and annotate the necessary data first – before model training could even begin – further intensifying the time crunch before the conference.
Partner Utopia needed to find their own partner: an outsourcing firm with proven AI expertise to resolve the training data issues. This partner’s development team would need to be quick, agile, and have skills precise enough to deliver a software product in 12 weeks to help Partner Utopia stand out at the conference and generate buzz ahead of release day.
Divelement is a nearshore technology consulting company that has been developing innovative AI-based software solutions since it started in 2020. They worked closely with Partner Utopia’s in-house team to create a tailored MVP development strategy that would achieve a successful launch in the 12-week window while simultaneously overcoming data management and model training challenges.
Divelement’s team knew they needed to use a tech stack that could deliver a production-grade MVP on the double. They used proven solutions like NextJS deployed to Vercel to simplify the initial project set-up and to provide the CI/CD (continuous integration/continuous delivery) dependencies that they’d otherwise have to build themselves. Once they selected those technologies, the team chose Clerk for authentication because it works seamlessly with NextJS.
Divelement went with PostgreSQL for the application database and the vector database because it’s the most powerful open-source database technology on the market. PostgreSQL also supports both traditional and AI-based applications, making it the perfect scaffolding for the app layer as well as the RAG (retrieval-augmented generation) architecture.
The team chose Python and FastAPI because they make it easy to build AI inference endpoints quickly. Render, like Vercel, eliminates complexity in deploying AI APIs to production. Finally, Divelement wanted to work with OpenAI-based models for fine-tuning because they support the use of synthetic data to supplement the limited initial data provided for model training.
Divelement also took multiple training approaches to fine-tune the models. The team employed a human-based feedback loop (also known as RLHF, or reinforcement learning from human feedback) for quantitative analysis. They also used a technique called SPIN (self-play fine-tuning) which allows an AI model to improve itself by playing against its previous iterations. This combo of training techniques allowed them to start with human-annotated data and then move on to automated fine-tuning, so they could create strong models that met with Partner Utopia’s stringent requirements.
Divelement completed the project on time and on budget, allowing Partner Utopia to launch its groundbreaking product as planned. Thanks to an innovative, fully functional MVP, Partner Utopia was able to sign the first 75 users at the conference launch event.
Divelement enabled Partner Utopia to meet an accelerated deadline while saving on development costs by leveraging their nearshore operating model, efficient operating practices, and diverse expertise. As an added bonus, engaging with Divelement saved Partner Utopia from having to invest significant amounts of time and money in recruiting and onboarding its own AI experts.
Partner Utopia lacked the staffing and expertise to build its MVP with existing resources in the short time allotted. Outsourcing the project with Divelement provided valuable benefits, including:
Partner Utopia needed resources with expertise in a wide variety of tools, languages, and AI technologies, but recruiting or training for these specific skills would cost more money and create unnecessary delays. Divelement had staff on hand with the exact expertise required for the MVP build, which allowed them to hit the ground running.
On average, outsourced developers cost less per hour than full-time employees. Companies pay reduced employment tax rates and aren’t responsible for the outsourced developer’s benefits, training, or medical leave.
Outsourcing to Divelement’s experts allowed Partner Utopia to present its MVP within the 12-week timeframe. Divelement’s team uses agile best practices like iterative development and continuous testing to reduce delivery cycles without sacrificing quality.
Divelement is a nearshore technology consulting company that helps businesses solve problems through technology and accelerate their digital transformations. Our approach starts by fully understanding what a client is trying to accomplish, determining if technology is the right way to achieve it, and then ensuring we have clear client outcomes to work backward from to build a successful roadmap.
Divelement’s team is based in Mexico. Our proximity to the U.S. offers numerous advantages over offshore teams, including those listed below.
Developers have a high level of English fluency, reducing the risk of miscommunication.
Mexico has a similar workplace culture, reducing friction over social faux pas and other mix-ups.
Overlapping time zones between U.S. offices and Divelement’s team facilitate real-time collaboration.
Divelement’s team offers the same (or higher) expertise as onshore dev teams but at a reduced cost.
Divelement has a proven track record of successful projects and happy clients, and we provide complete transparency into our staffing and development practices.
96%
Referral Rate
1M+
Monthly Users
150+
Projects Completed
60+
Happy Clients
135k+
Hours Worked
60+
Team Members
Divelement has the expertise to handle every stage of your project lifecycle. We provide team augmentation to fill specific knowledge gaps, managed services for long-term maintenance and support, and full onshore consulting for close collaboration on mission-critical initiatives. We help our clients meet their goals faster without the expense or difficulties of hiring in-house experts.
Minimum viable product (MVP): A basic version of a software product that has the functionality required to address a particular issue for early adopters. Its principal goal is to get input from early adopters to validate your idea and direct further development. Learn more.
NextJS: An open-source development framework by Vercel that provides server-side rendering and static page generation for React-based applications.
Vercel: A cloud-based platform as a service (PaaS) built around a composable architecture. Vercel also maintains the NextJS dev framework.
CI/CD (continuous integration/continuous delivery): A set of tools and practices used in many DevOps environments to automate key code integration and delivery processes, such as code builds, integration testing, unit testing, regression testing, feature deployment, and infrastructure provisioning.
Clerk: A user authentication solution that works seamlessly with NextJS and AI applications to deliver capabilities like single sign-on (SSO) and multi-factor authentication (MFA).
PostgreSQL: A free, open-source relational database management system. PostgreSQL is known for its reliability, robust feature set, extensibility, and active developer community.
Retrieval-augmented generation (RAG): A generative AI technology that uses both the model’s training data and information retrieved from outside sources to provide more accurate and natural-sounding results.
Python: One of the most popular general-purpose programming languages, used to develop a wide variety of applications for enterprise, web, mobile, AI, and more. Python is widely used due to its versatility and its comprehensive library that streamlines the coding process.
FastAPI: A web development framework that makes it easier and faster (hence the name) to build Python APIs.
Render: A cloud application development platform that automates and streamlines the building, deployment, and scaling of cloud-native apps.
OpenAI: The makers of some of the most widely used foundational large language models (LLMs) used for generative AI, known as GPT.
Synthetic data: Artificial data that looks to AI models like real information. Synthetic data can be used to make a training dataset more robust without the time and expense of collecting and annotating real data. It can also replace sensitive data like personally identifiable information (PII) and confidential company records to mitigate privacy risks.
Reinforcement learning from human feedback (RLHF): An AI training technique that has human engineers supervise the first stage of training or fine-tuning to establish a baseline of real human preferences and feedback. The human feedback is then incorporated into the model’s training algorithm so it can self-learn more effectively.
Self-play fine-tuning (SPIN): An AI fine-tuning technique that takes place after supervised (i.e., RLHF) fine-tuning has already occurred. SPIN has an AI model play against previous iterations of itself by attempting to discern between human-generated responses and those generated by the older iteration. SPIN enables an AI model to enhance its own capabilities without requiring additional human-annotated training data.
Need some guidance with your software development project?
Look no further! We’re offering a free consultation, to help get you moving in the right direction.
