May 2026/AI/ML/professional
EduAccess Agent ID
Most AI bots in education are basically just glorified slot machines. You ask about scholarships and pray they don't hallucinate a fake deadline. I built EduAccess because I was tired of that. Instead of a chatty bot that spits out endless paragraphs of unreliable advice, I wanted an agent that forces the AI to be completely transparent about its limitations and strict with its outputs.
To actually make that happen, I completely stripped away the chat interface and went API-first. I built a rigid Planner-Retriever-Reasoner pipeline using Python and FastAPI. The real secret sauce here isn't the LLM prompt. It's Pydantic. It ruthlessly validates every single output against a strict JSON schema. If the AI forgets to return a confidence score or tries to hallucinate a requirement, the response gets dropped instantly and fallback logic kicks in.
Moving away from unstructured text to strict JSON contracts changed how I look at AI development. It taught me that when building for high-stakes domains like financial aid, safety and traceability matter way more than just sounding conversational. The end result? An agent that doesn't just guess, but actually calculates.