Case Study: Building Shikhi AI — An AI Learning Platform That Placed Top 5 Nationally

Shikhi AI is our own AI-powered learning companion — it turns a student's study material into personalized lessons, quizzes, and instant conversational answers, and it placed Top 5 nationally at the bdapps Innovation Summit 2025. Here's how we approached it end to end, and what it taught us about building AI products that actually help the people using them.

The problem we set out to solve

Students often have the material they need to study, but no one to ask the moment they get stuck. Personalized tutoring is effective — and expensive, and doesn't scale to the number of students who need it. We wanted to know: could an AI companion close that gap without feeling like a generic chatbot bolted onto a study app?

What we built

• Personalized lessons generated directly from a student's own study material — not a generic curriculum
• Quizzes that adapt to what the student has actually covered
• Instant, conversational Q&A that feels like asking a tutor, not searching a help center

The frontend is built on Next.js, with an LLM integration powering the conversational and lesson-generation layer.

The technical decision that mattered most

The obvious failure mode for any AI tutor is confidently wrong answers. We grounded every response in the student's actual study material — the same retrieval-based approach we cover in our AI solutions guide — rather than letting the model improvise from general knowledge. For an education product specifically, an answer that's fluent but wrong is worse than no answer at all, so accuracy took priority over how impressive the demo looked in early iterations.

How we worked

We built Shikhi AI the same way we build for clients: weekly internal demos, a frozen core scope for the first version, and a deliberate decision to prove the conversational Q&A worked well before expanding into adjacent features. Building our own product with our own process is the best test of whether that process actually holds up.

The recognition

Shikhi AI placed Top 5 nationally at the bdapps Innovation Summit 2025, hosted by Robi/bdapps — competing against startups from across Bangladesh. It's the proof point we point to when clients ask whether we can build serious AI products, not just integrate an API and call it done.

Where the project stands

Shikhi AI is still in active development and not yet publicly launched — we'd rather say that plainly than overstate where it is. What we can say: the lessons from building it — grounding, honest scoping, and treating an AI feature as a product decision rather than a technical flex — now shape every AI project we build for clients, including the guardrails covered in our AI agents guide.