Hire — AI & GenAI
GenAI that people actually keep using.
This is the domain we lean into hardest — SaiyamAI, our own GenAI product, is live in both stores, which means the trade-offs in this section aren't secondhand. AI/ML engineers here work on LLM-backed features, model selection and integration, and the unglamorous data-handling decisions that determine whether an AI feature is trustworthy or just a demo.
What a typical AI/ML engagement looks like
- —An engineer joins to add a GenAI feature to an existing product — chat, summarization, personalization — scoped against your actual data and users
- —Or a from-scratch AI/GenAI product build, design through release, the same Build shape as any other end-to-end engagement
- —Model selection and integration work: which third-party LLM API fits the latency/cost/quality trade-off for your use case, evaluated rather than assumed
- —Privacy and data-handling decisions made explicit and written down, not glossed over because 'it's just AI'
Stack & product considerations
- —Third-party LLM APIs vs. a fine-tuned or custom model — most product use cases don't need the second, and we'll say so rather than sell the bigger build
- —Python for the ML/data layer, integrated into whatever backend and mobile stack the rest of the product runs on
- —What data actually needs to leave the device or get logged at all — our own product's stance ('collects nothing') is a real design decision, not a slogan, and we can walk through why
- —Latency and cost budgets that make an AI feature usable at your actual scale, not just in a demo with no traffic
Questions people ask.
Adding a GenAI feature to an existing product, building a new AI-native product end to end, or advising on model/architecture choices before you commit engineering time. SaiyamAI — our own screen-time app with a GenAI conversational core — is the proof this isn't theoretical for us.
Almost always the former — a well-chosen third-party API gets most products to a better outcome faster and cheaper than a custom-trained model, and we'll tell you plainly if your use case is the exception rather than default to the bigger, more billable build.
As a design decision made up front, not a compliance afterthought. Our own product's approach — collecting nothing it doesn't need — reflects how we'd default on client work too: minimise what's logged or sent to a third-party model, and say clearly what does leave the device when something must.