AIVETech · AI Feature Sprint · Our work

Explore our work.

Founder-built software and a clearly labelled sample. Evidence you can inspect before discussing a project.

Discuss Your Project →

Founder-built SaaS · not paid client work

Soil — customer commitments through delivery

Soil is built by Kartik Malhotra. It helps B2B software teams record a customer promise, assign an owner and deadline, review the work behind it and separately confirm customer delivery.

The product problem

A completed engineering task does not prove a customer received what was promised. Teams need a commitment record that stays visible through review and delivery confirmation.

Implemented capabilities

Technical approach

Python and FastAPI serve the application. Jinja and htmx provide the interface; PostgreSQL holds workspace and workflow records. Rules handle core commitment tracking. Optional model-assisted answers and drafts sit alongside the workflow.

  1. Record or reviewA manual promise or a candidate from an authorised source.
  2. Check evidenceOwner, deadline and explicitly linked work within access boundaries.
  3. Confirm deliveryA person reviews evidence and records what reached the customer.

What you can inspect

The public product site and its fictional commitment demo are available without connecting your tools. The source implementation and deployed version were reviewed for this description.

Current limits

The public demo uses fictional data. Connected evidence depends on provider access and configuration; not every advertised workspace view is the same as a verified end-to-end integration. Core commitment tracking does not require an AI model. This page makes no claim about customer adoption, paid-client outcomes, revenue or production performance.

Visit Soil ↗ · Discuss an integration or workflow like this →

Scripted browser demonstration

Clinic receptionist sample

Problem: a clinic front desk repeats answers about hours, fees and appointment requests.

Implemented: a browser chat recognises a small set of phrases and displays fixed responses using JavaScript. No message is sent to an AI provider or stored by the sample.

Limits: fictional clinic and appointment data; no voice, calendar, WhatsApp or staff integration; no real bookings or clinical advice. It is not a paid-client deployment or evidence of live voice engineering.

Try the sample on the homepage →

A connected implementation would need agreed channel providers, a booking API, consent and retention decisions, escalation paths and representative tests.

Engineering decisions, explained

The engineering notes cover agent guardrails, retrieval evaluation and voice latency. They explain design considerations; they are not client case studies.

Bring us one product workflow

Tell us what your users need to do, where the data lives and what a successful feature would look like. We will review fit before agreeing any paid implementation.

Discuss Your Project →