Participated in public demo 2606 4W with TrueRead

The problem [00:00] Parents preparing kids for competitive secondary-school entrance exams in the UK are flooded with workbooks, tutors, and practice apps that hand back a bare score, never revealing what their child was actually thinking or what to do next when an answer is wrong. What they built [00:45] He built TrueRead: one system with two windows onto the same pipeline, a story-world game the child plays, and a companion app the parent checks. Every answer, including the specific wrong option a child picked (not just right/wrong), feeds a deterministic engine that turns raw answers into a tiered, evidence-grounded read of a child's reasoning. Instead of a bare topic score, the parent app names the precise misconception behind a mistake and turns it into one clear next step. Charit built the product with heavy emphasis on child-data safety: all 24 identified risks, including all seven zero-tolerance red lines, were mitigated before launch, with a child's name never reaching the underlying LLM in raw form. What happened [02:07] Item-level generation passed strict quality checks 79% of the time, with a 4.31/5 mean quality score after a mandatory critic gate, materially stronger than raw, ungated generation, which showed roughly 37% filler on the hardest question constructs. The underlying engine passed 6,979 automated tests, independently re-verified through an adversarial commit gate. The ask Charit is looking for help reaching the right audience with the right message for TrueRead.