# Indagora - **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon) - **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST) - **Where:** Online - **Team:** [Alejandro Hernandez Vasquez](https://cerebralvalley.ai/u/experienciapcol) - **GitHub:** https://github.com/experienciapcol-sudo/indagora - **Demo video:** https://youtu.be/G1YpbEGqHDs - **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery - **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3477 ndagora is an organizational diagnosis agent for management consultants. It investigates before it prescribes. The problem: a consultant interviews four people and gets four opinions. The client wants answers by Friday, and the easy move is to recommend something before anything has been checked. In our demo case, a fictional furniture workshop, the owner says "the team is lazy." The evidence shows a new hardware supplier that takes 9.4 days instead of 3.1. How it works. The consultant pastes interview notes or transcripts, or records audio, and marks who is speaking. Every sentence gets an ID. Specialist model roles read each source in parallel and propose claims by pointing to sentence IDs. The model never writes a quote: the code rebuilds it and rejects pointers to the interviewer's questions, anchors that are not in the sentence, and figures the quote does not contain. Evidence levels N0 to N4 are computed by code. One person's word never rises above N2; two different roles that agree, plus the consultant's click, reach N3; a document with the data reaches N4. A goal-directed agent then applies a published method chosen by a rule in the code: Kepner and Tregoe's problem analysis (deviation, Is / Is Not, cause testing) for an acute problem, Weisbord's six boxes for a chronic one, with an Ishikawa diagram. Before every move, code guards approve or reject it, and the full trace is visible. A cause below N3 only admits verification actions. The agent declares a principal cause only when its evidence holds; otherwise it returns a partial delivery that says exactly what the consultant needs to provide. The output is work a consultant can hand to a client: a verification plan (PDSA), an intervention plan sorted by effort and impact, SMART indicator sheets with a baseline, a Pareto built only from documented figures, a guide for the next interview, and a feedback report. Every line links back to the sentence behind it. We tested it on seven fictional cases, three runs each. The agent made 0 premature interventions in 21 runs. A single prompt with the same model proposed 2.0 to 3.3 per run, none of them checked against evidence. A Laboratory mode lets anyone practice: it generates a fictional company with a hidden cause and simulated employees who only know their own facts. Live demo (replays recorded runs at no cost): https://indagora.experienciapcol.workers.dev All data in the project is fictional. ## More from Meta Global AI Developer Hackathon - [Binary Pulse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3474) - [Rajay McPherson](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3475) - [Monster Paws](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3476) - [Jotika](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3478) - [koxy Builds](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3479) - [Falden](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3480) --- Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3477. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt