Home » AI Lab

Don't Take Our Word for
It - Try the AI We've Built

The Stacknize AI Lab is a live showcase of AI capabilities we have built and deployed. Not slides. Not bullet points about what we ‘can do’. Actual working tools — some running in production on client platforms, some built specifically for this page — that you can interact with right now.

Demo 1 — Medical Consultation Summary (MedGemma)

Paste or upload a sample doctor-patient consultation transcript. The AI analyses the conversation using Google MedGemma — a medically specialised LLM — and generates two outputs:

Agency Implementation Notes

Build as a two-panel interface: left panel = text input area (paste transcript) or file upload (PDF/TXT); right panel = output rendered in two labelled sections: 'Clinical Summary (Doctor View)' and 'Patient Summary'
Backend: FastAPI endpoint calling MedGemma via Google Cloud Vertex AI; stream the response for perceived speed
Add a sample transcript button so visitors can try it immediately without needing their own content
Show a small badge: 'Powered by Google MedGemma — Medically Specialised LLM'

Demo 2 — AI Call Centre Agent

A 3–5 minute video demonstration of the AI Call Centre Agent handling a real inbound customer enquiry scenario. The video shows the full journey: call arrives via SIP, AI greets the caller naturally, handles a multi-turn conversation, retrieves relevant account information from a demo CRM, and resolves the query — all without human intervention.

What the Video Should Show

A realistic customer scenario — e.g. a telecom subscriber calling to check their data balance and upgrade their plan
Natural conversational pacing — demonstrating the sub-1.5-second response latency between caller and AI
The agent dashboard in the background — showing the live call log, intent classification, and confidence scores in real time
A human handoff scenario — the caller asks something out of scope; the AI escalates cleanly with a warm transfer message

Agency Implementation Notes

Demo 3 — Document Q&A (RAG Pipeline)

Upload any PDF document and ask questions about it in natural language. The system uses our RAG pipeline — Retrieval-Augmented Generation — to extract relevant passages from the document and generate accurate, grounded answers. Answers include source references so you can verify exactly where the information came from.

Why This Matters for Prospects

This demo answers the most common AI question we get from enterprise clients: ‘Can your AI answer questions about our internal documents, policies, and data — without hallucinating?’ This demo proves it, live, with the visitor’s own document.

Agency Implementation Notes

File upload (PDF, up to 5MB) with a drag-and-drop interface; also offer 3 sample documents (e.g. a telecom product guide, a medical information sheet, a government policy document)
Show the RAG process visually: 'Searching document... Found 3 relevant sections... Generating answer...'
Display source excerpts below each answer with page reference — this is the trust signal that distinguishes RAG from hallucination
Backend: LangChain + pgvector; chunking and embedding on upload, retrieval and generation on query; FastAPI

Demo 4 — Voice-to-Text Pipeline

Upload or record a short voice clip and receive a structured transcription and summary. This demonstrates the speech-to-text processing pipeline that powers both our AI Call Centre Agent and the voice input capabilities in our mHealth platform.

Agency Implementation Notes

Option A (simpler): Video walkthrough showing a recorded demo — voice input, Whisper processing, structured transcript output
Option B (interactive): Browser microphone input using the Web Audio API; audio sent to a Whisper API endpoint; transcript displayed with timestamps and a one-paragraph summary
Show language and accent examples — demonstrating Whisper's cross-accent accuracy is a genuine differentiator for African and Asian market clients