AI Use & Limitations
Last updated: 9 September 2026
We build systems that use AI models. This page states plainly what that means, where the boundaries are, and what we will not claim. It applies to the chat widget on this site and describes the principles we apply to client work.
AI output can be wrong
Language models generate plausible text. They can state incorrect things confidently, including about pricing, capability and timelines. Treat anything the chat widget says as a starting point, not as a commitment from Kognisol. Anything that matters commercially should be confirmed in writing with a human.
Which providers process what
The chat widget on this site sends your message to Groq, which runs the model that generates the reply. We do not use your chat messages to train models. If a client engagement requires a different provider or a specific data-residency arrangement, that is agreed in the engagement.
Human accountability
We design automations with explicit approval points rather than unattended authority. In practice that means:
- Actions that spend money, contact a customer at scale, or change a record of consequence are designed to require a human decision.
- Systems are built to log what they did, so a person can reconstruct it afterwards.
- Where a model is uncertain, the intended behaviour is to escalate to a person, not to guess.
Where we will not let automation decide
For healthcare, finance, legal, employment and similar high-impact domains, our systems support the professional; they do not replace professional judgement. Specifically, we do not build automations intended to diagnose, to triage clinically, to give investment advice, to make hiring decisions, or to make legal determinations without a qualified human in the loop.
What we do not claim
- We do not claim regulatory certification we have not obtained.
- We do not present demonstrations as client outcomes.
- We do not guarantee that a model will never produce an incorrect output, because no one can.
Reporting a problem
If the chat widget or any system we built produces something harmful, misleading, or wrong, tell us at nithin@kognisol.in and we will look into it.