One week.Real AI problems.One working build.

Brands, marketers and creators need to post regularly on Instagram, LinkedIn and X, but every good post or video still takes hours of scripting, designing, editing and repurposing. Small teams can't keep up, so content goes out late, rarely, or looks generic.

Build an AI system that takes an idea, product or brand brief and turns it into publish-ready marketing content in any format: video, audio, text or visuals. It should fit each platform's format and style, match the brand's or creator's voice, and be good enough to post with little or no editing, cutting the time from idea to published content from hours to minutes.

Most ways to learn AI treat every learner the same. A beginner and an experienced developer get the same videos, in the same order, at the same pace, whether they are struggling or bored. Nothing notices what a learner already knows, where they get stuck or how they learn best, so people drop off or learn things they never use.

Build an AI tutor that teaches AI and personalises itself to each learner. It should work out what the learner already knows and wants to achieve, build a learning path around that, and keep adapting as they go: changing difficulty, pace, explanations and practice tasks based on how they perform. Learners should learn by doing, and the tutor should show that each one is actually improving.

Companies sit on large amounts of business data, such as CRM records, sales pipelines, customer lists and spreadsheets, but getting a clear, accurate answer from it is hard. The data is scattered, duplicated and often out of date, so teams spend hours pulling reports and still end up deciding on gut feel.

Build an AI system that works on a company's business data and turns it into accurate answers and decisions. Pick a company persona and a task. For example: you are a sales team with thousands of leads in your CRM. The system should find the most relevant leads, explain why each one matters, and decide which ones the team should reach out to first. Every answer and decision should trace back to the data behind it, with a human approving the final action.

Most AI products still expect people to type. But many real situations, such as a job interview, a support call or explaining a problem you can see, work far better by speaking, showing or sharing a screen. In these moments a text box feels slow and unnatural.

Build an AI experience where people interact through voice, images, video or the screen itself, and where that interaction clearly works better than text would. It should hold a natural, real-time conversation: respond quickly, handle interruptions, understand what it hears and sees, and complete the task, such as running an interview or resolving a customer's issue.

Most AI products are built English-first, while a large share of India is more comfortable in Hindi, Tamil, Telugu, Bengali, Marathi or another local language. Millions of people are left out of what AI can already do.

This one is wide open. Build anything for India that works in Indian local languages: any domain, any user, any format. Pick a real problem, the people who face it, and build an AI solution they can use in their own language.