Skip to content
Software Muskets

Fig 2.0Services · Generative AI

Generative AIAssistants and copilots your users will actually keep open.

We design and build generative AI products: customer and employee assistants, in-product copilots, document and content pipelines, and voice interfaces. We start with the job the user is trying to finish, pick the model after measuring, and write prompts the way we write code: versioned, tested and reviewed.

Engagement models

  • Discovery sprint (2 weeks)
  • Build and launch (8–12 weeks)
  • Dedicated squad

Fig 2.1What you get

What you walk away with.

  • 01

    Assistants

    Customer-facing and internal assistants with grounding, memory and escalation to humans.

  • 02

    In-product copilots

    Features inside your product: draft this, summarise that, explain this number, fix this form.

  • 03

    Document pipelines

    Extraction, classification and summarisation over invoices, contracts, reports and forms at scale.

  • 04

    Voice

    Speech in, speech out, in Indian languages, for users who would rather talk than type.

  • 05

    Prompt engineering as engineering

    Prompts in version control with tests, reviews and rollbacks.

  • 06

    Model strategy

    Which model where, at what cost, with what fallbacks. Written down and revisited quarterly.

Fig 2.2Proof

Where we have done this.

  • 01

    On-device voice search shipped to first-time smartphone users in Hindi.

  • 02

    Content and translation pipelines that generated every string of a bilingual app from one source.

Fig 2.3Stack

Tools we reach for.

Chosen per project on fit and cost, never on habit. We will tell you when something simpler will do.

  • Claude, GPT, Gemini, open-weight models
  • Speech recognition and synthesis
  • TypeScript, Python
  • Streaming UIs in Next.js and Flutter

Fig 2.4Questions

Asked often.

How do you control cost?

Prompt caching, smaller models for simpler steps, batching for offline work, budgets per feature and a dashboard that shows spend per user action.

Do we own the prompts and evals?

Yes. Everything is in your repository at the end.

Have a product in mind?
Let's build it.

Tell us what you are trying to make happen. You will hear back from an engineer within one working day.