The Challenge

The client, an EdTech company focused on social emotional learning, came to us with an ambitious roadmap of AI models for social emotional learning. The initial product strategy leaned toward an agentic architecture built on hosted, third-party commercial LLMs. However, relying entirely on external API endpoints posed significant IP risks and created long-term dependency on third-party pricing and infrastructure, diluting the unique value of the client’s proprietary data and domain knowledge.

The Choice

Rather than taking the easy route of wrapping third-party APIs, we challenged the brief. We convinced the client to invest in building their own Small Language Model (SLM). This strategic shift ensured they retained complete ownership of their intellectual property, reduced long-term operational costs, and allowed them to leverage their unique data assets across multiple customized personas.

What We Built

We engineered an end-to-end proprietary AI platform and retraining pipeline:

  • Custom SLM Architecture: Developed and trained an initial core model, then fine-tuned it across multiple educational personas using the client’s dataset.
  • Automated Retraining Pipeline: Built a scheduled pipeline that ingests new data, system exceptions, and edge cases to continuously train and deploy updated model versions.
  • Agentic Orchestration & Tracing: Integrated LangGraph for multi-agent workflows, combined with vector databases and open-source OCR engines for document ingestion and tracing.
  • Full Technical Handover: Designed the retraining pipeline to be fully operated and managed by the client’s team without vendor lock-in.

The Outcomes

  • Tech Stack: Agent Orchestration Framework (LangGraph), Public Cloud, Open-Source OCR Engine + SLM for OCR, LLM Tracing & Evaluation Frameworks, Vector Database.
  • Deployment Model: Scalable public cloud environment hosting custom SLM inference endpoints and automated pipeline triggers.
  • Client Trust & Partnership: This marked the foundation of our ongoing partnership with this client—reflecting Rite Technologies' commitment to long-term collaboration, where most of our partners rely on us for between 5 and 10 years to expand their core AI and software architectures.

“In technology development, where things don’t always go as planned, having a team you can trust, communicate openly with, and count on makes all the difference.”

— CEO of the company, InnerJoyed

The Handover

By moving away from third-party model dependencies, the client secured full ownership over their AI IP and core tech stack. Following the successful deployment, the client secured fresh funding and engaged us to build their next generation of AI models.

Tech Stack

AI
Small Language Model
SLM
Lang Graph
Agentic AI
Vector Database
LLM Tracing
Social Emotional Learning
OCR
Public Cloud
IP Ownership

If your product is currently relying on external AI models and you want to build and own your proprietary solution,

let's connect.