AI-Powered Veterinary Forensic e-Learning
A specialized TVET platform using RAG (Retrieval-Augmented Generation) to train veterinary forensic professionals.
This is probably the biggest project I’ve worked on so far. I'm building it for an organization to help train veterinary forensic professionals in TVET programs. It’s been a massive learning curve because it’s not just about the code; it’s about making sure the AI is accurate enough for medical and legal standards.
Anchoring the AI with RAG
One of the big things I wanted to solve was AI "hallucinations." In veterinary forensics, you can't have an AI just making things up. I used Retrieval-Augmented Generation (RAG) to anchor the AI’s answers strictly to vetted forensic manuals and documents.
I’ve been working on a pipeline that takes technical PDFs, chunks them up, and stores them in a Vector Database. When a student asks a question, the system finds the exact right paragraph from the manual to help the AI answer.
More than just a Chatbot
The system is more than just a chat window. I had to build a proper RBAC (Role-Based Access Control) system for Admins, Trainers, and Trainees, which was my first time dealing with such complex permissions. I also built a "Competency Mapper" that takes quiz results and maps them to specific academic units, so trainers can see exactly what skills students are missing.
Why this is hard
The hardest part hasn't been the backend, but actually tuning the RAG model. Getting it to prioritize forensic standards over general internet knowledge took a lot of trial and error with prompt engineering and different chunking strategies. I also had to think a lot about data privacy, making sure the organization’s data stays secure.
It’s been a huge experience learning how to build a production-grade AI system from the ground up, even if the repo itself is private.