An AI mobile training platform for SANA Hotels

Twistag built an AI training platform for SANA Hotels that turns the group's existing PowerPoint and Word library into three formats from a single document upload: interactive text courses with quizzes, video lessons delivered by AI-generated avatars, and personalised learning paths scoped by role, property and language. SANA had been flying trainers between hotels every month to deliver the same decks in person, and every new property added a stop on the circuit. The platform runs serverless on AWS with Amazon Bedrock and Anthropic Claude, and was designed to absorb SANA's projected 20% workforce growth to 2027 without adding training staff.

SANA's training model was small-team and hands-on. Trainers travelled from hotel to hotel every month delivering sessions in person from PowerPoint decks. As the portfolio expanded, the arithmetic stopped working: every new property added a stop on the circuit, every new hire triggered another in-person onboarding session, and every policy update required re-travelling to every location. There was a quieter consistency problem too — a trainer covering the same material for the fifteenth time in a month adjusts, abbreviates and skips, so new hires at different properties received subtly different versions of the same training. And SANA's library of hundreds of PowerPoint decks and Word documents represented years of institutional knowledge in a format that made it unusable: no review between shifts, no comprehension check, no completion tracking, and manual effort for every translation.

Twistag built a platform that converts SANA's existing library into three output formats from a single document upload, running entirely on serverless AWS infrastructure with Amazon Bedrock and Anthropic Claude doing the AI work. The platform separates into three layers that evolve independently: content ingestion and generation, the learning platform employees progress through, and the personalisation layer deciding which modules each employee sees.

Converting a corporate PowerPoint into a learning module is not a single model call. The pipeline first extracts document structure — headings, bullets, images, speaker notes, embedded media — with platform-specific extraction logic written separately for PowerPoint and Word, because the structural conventions differ and real corporate files are messy. Structured content then passes through Bedrock with Claude as a sequence of discrete steps rather than one combined prompt: identify learning objectives, restructure into a mobile lesson flow, generate scenario-based quiz questions testing application rather than memorisation, and produce metadata tags for personalisation. Running these sequentially produces more reliable output, because each step receives clean structured input from the previous one.

One upload, three surfaces, and a path that feels personal. Text courses are five-to-ten-minute modules sized for a break between shifts. Video courses use AI-generated avatars where tone and delivery matter as much as the words. The personalisation engine assembles each employee's path filtered by role, property and language: the same fire-safety document produces one version emphasising evacuation communication for guest-facing staff and another emphasising equipment shutdown for maintenance, from a single upload.

Stack: Amazon Bedrock, Anthropic Claude, serverless AWS. Services: AI agents and applied AI, AI-native product engineering, content generation pipelines.

Training that previously required scheduling and a train ticket now reaches every employee at every property within minutes. The training team's time has shifted from logistics to content development, comprehension is tracked, and languages are handled without per-document manual translation. Hundreds of legacy training documents have been converted into mobile courses. The platform was designed to absorb SANA's projected 20% workforce growth to 2027 without adding training staff — same team, same content engine, more properties feeding through it. Course completion is tracked for the first time, so the training team can see which modules land and which need rewriting rather than guessing from the room.

Trainers off the road. Content in pockets. Comprehension tracked. Languages handled.