Founding AI Systems Architect - LLM Council & Private GPT (In-Residence)

Founding AI Systems Architect - LLM Council & Private GPT (In-Residence)

Company:

Aonxi Location:

In-Residence (North India, Himalayas) for first 3 months, then remote Type:

Full-time, founding role

What we’re building Aonxi is building a

private AI brain per business -

a ChatGPT-like interface powered by a

Language Model Council

that reasons, judges, learns from outcomes, and closes the loop from

conversation → decision → action → revenue .

This is not a chatbot. This is

decision infrastructure .

Your mission Design and build the

Aonxi Council : Multiple LLMs generate independent answers Multiple LLMs judge those answers using structured rubrics A winning answer is selected with confidence and risk flags Each client has a

private AI brain Real outcomes feed back into the system to improve decisions over time

What you must deeply understand

Multi-LLM orchestration and arbitration Judge/respondent separation Bias control and evaluation design Private memory per tenant Feedback loops tied to real outcomes Safe AI decision systems

What you’ll build

A ChatGPT-like UI for employees and clients A multi-model reasoning backend (LLM Council) Private per-client AI memory Outcome-aware learning loops Guardrails for high-stakes decisions

In-Residence requirement For the first

3 months , you will live and work with the Aonxi core team in a co-living, co-working setup in the Himalayas (near Shimla). Accommodation and food are provided. This proximity is essential to build the system correctly.

Who should not apply

Prompt engineers Tool integrators “AI enthusiasts” Anyone who has never designed a decision system

How to apply - Answer 4 questions. Email

origin@aonxi.com

with: A short explanation of how you’d design an LLM Council from first principles One system you’ve built that learned from outcomes Confirmation you can do 3 months in-residence You are running a private AI brain for 1,000 businesses. One AI decision causes wrong emails to be sent to 200 clients. No data is corrupted. The models behaved “as designed.\" From first principles, what failed - and what ONE structural change would you make so this can never happen again? Subject:

Founding AI Systems Architect