Meera

Every university help desk runs on the same broken loop: students send vague urgent emails, staff ask for missing details, and actually important cases sit in the queue. For the KPMG Academic Innovation Challenge, we built Meera as an agentic AI front door that tries to resolve the problem first and only escalates when a human is genuinely needed.

Students never pick a department. They describe the issue naturally, and Meera routes it across IT, Registrar, Finance, Health Services, or Student Services. If a human is needed, staff receive a structured case with summary, collected information, missing details, suggested next steps, confidence score, and an audit trail. The LLM proposes a department, then thresholded math over a knowledge graph confirms, corrects, or escalates the route.

We also built Battle Mode, a game-like interface over the exact same agent and ticket pipeline. It proved the architecture was flexible: once the underlying case model is solid, the user experience can change without redesigning the engine underneath.

The desktop version was the part I was proudest of. Meera can draw transparent click-through overlays over any app on screen, using a vision model plus a zoom-refine second pass to locate the right UI target more reliably. It feels closer to someone pointing at your screen than a normal chatbot.

Under the hood, it is one Next.js codebase running in the browser, Electron desktop, and Cloudflare Workers through runtime-selected adapters. Voice uses Whisper for speech-in and Deepgram Aura for speech-out. Model calls stay server-side, and the system never approves payments, changes records, or makes medical decisions. It gives people a head start while leaving real decisions to humans.