.png&w=256&q=75)
@roguejuanservices68394
2
Events attended
2
Submissions made
Looking for experience!
Socials
🤓 Latest Submissions
EnquiryOS Shadow Audit
Customer enquiries are often buried, partially answered, forgotten after follow-ups, or lost in incomplete message histories. Traditional CRMs record activity, but they do not reliably prove which conversations still need attention. EnquiryOS audits historical conversations through a live Evidence Tribunal. Each conversation is investigated, challenged against the available evidence, and classified as a Verified Attention Item, Needs Human Review, or Cleared After Evidence Review. What makes EnquiryOS different is that the AI does not simply produce a score or recommendation. Every actionable finding can be traced back to exact source-message evidence through Evidence Lineage. The Tribunal Ledger also exposes the conversations the AI deliberately chose not to escalate, making rejection and uncertainty visible rather than hiding them. The Owner Action Brief converts the audit into a business-ready action view: what can be recovered now, what must be verified first, what was safely cleared, which leakage patterns were detected, and which conversations should be handled next. For verified missed enquiries, the Human Recovery Studio prepares an editable, evidence-grounded Recovery Brief containing supported facts, facts still requiring confirmation, risk checks, source-message provenance, and a human approval checklist. For uncertain cases, EnquiryOS deliberately stops. If the conversation history is incomplete, the system returns REVIEW_REQUIRED and does not generate a send-ready customer reply. Nothing is sent automatically. The hackathon demonstration uses seven synthetic conversations for Lumina Events. The live Evidence Tribunal identifies two verified missed enquiries, one case requiring human review, and safely clears four conversations that do not require recovery. EnquiryOS is designed as an evidence-backed revenue safety net: detect what was missed, prove why it matters, prioritize the right action, and recover only with human approval.
10 Aug 2026
ColdChain Sentinel
ColdChain Sentinel is a synthetic-only, advisory-only cold-chain intelligence platform that helps operators understand what may be going wrong and what a human should inspect first. Instead of treating every temperature excursion as a simple threshold alert, it evaluates data quality, sensor agreement, mapping uncertainty, and risk evidence before presenting a likely fault pattern and an inspection recommendation. Its Sentinel Thermal Behavior Learner (STBL) was trained offline on AMD GPU infrastructure using 171,000 synthetic training windows, 38 project-defined fault classes, and 19 weighted features. The neural model achieved 95.51% synthetic fault accuracy and 99.52% synthetic behavior accuracy. For the live demo, that research was distilled into transparent Python runtime logic, achieving 77.25% synthetic fault accuracy and 94.06% synthetic behavior accuracy without requiring a GPU, PyTorch, notebook, database, or external service at startup. The containerized application includes a command center, end-to-end case walkthroughs, an algorithm evidence console, a fault atlas, inspection guidance, and judge-ready validation evidence. Fireworks AI is available only as an optional, safety-gated explanation layer. Deterministic rules remain authoritative, provider output cannot trigger operational actions, and every recommendation requires human review. The product vision is an evidence and inspection layer for temperature-sensitive logistics. It could help quality, operations, and insurance teams interpret noisy or conflicting sensor evidence faster. The current project uses no real shipment, customer, pharmaceutical, patient, or sensor data. Controlled external validation and data-governance review would be required before any pilot.
13 Jul 2026
