Building an AI Coach That Helps You Take Action

The Challenge

Unlimits was already an AI-capable app when they came to us. Their platform lets users describe their dreams and goals in plain language, and turns that into a structured plan with daily micro-tasks — a personal coach, habit system, and task engine in one app that adapts to each user over time.

What's Inside Unlimits

As they scaled, three operational challenges became clear:

Quality assurance — almost everything the user sees is generated by AI, so keeping outputs consistently useful and relevant needed more structure than the original setup allowed.

Cost per user — every interaction had a real running cost attached, and the team needed clear visibility into what each part of the experience was costing them.

Visibility — as the experience got richer, the team needed to see exactly what happened inside any given user session, replay edge cases, and catch issues before users did.

What We Built

Optomize rebuilt the Unlimits AI engine from the ground up — giving the team clear visibility into every user interaction, a memory system that genuinely scales, a self-tuning loop that quietly improves the experience for each individual user, smart cost controls, and a platform built to grow. The result is an app that gets sharper the longer someone uses it, costs less to run, and is ready for the next stage.

Platform Capabilities

Clear Visibility

The team can now see exactly what happened in any user session — what the app did, how long it took, and what it cost — for any user, at any time. When something doesn't land right, they can see why immediately, instead of spending days digging through logs.

LangSmith trace dashboard

Every user session is captured in detail. When a response misses the mark, the team can see exactly why — no guessing, no two-day debugging exercises.

Self-Healing AI

The platform doesn't just respond — it learns. Every AI response is checked against how the user actually engages with it, the feedback they give, and their longer history on the app. Over time, the system quietly tunes itself for each individual user, so what they get in month three feels noticeably more on-point than what they got in their first week. No manual retraining required — the experience improves itself, one user at a time.

Long-Term User Memory

Unlimits needs to know its users over months — their goals, their patterns, the gap between what they say they'll do and what they actually do. We built a memory system designed to hold that context reliably as the user base grows, keeping responses personal and relevant without runaway costs.

AI-Powered Research Engine

The app actively researches the internet on each user's behalf — surfacing guides, articles, and resources that match where they are on their journey right now. Instead of pulling from a fixed library, it pulls in fresh, relevant material in real time, so a user building a fitness habit gets different reading than someone planning a career pivot — and what they're shown keeps evolving as they do.

Research engine in-app showcase

Personalised Notifications

Instead of blanket push notifications on a fixed schedule, the app now sends personalised nudges to each user — timed to when they're most likely to engage, and framed around what's actually relevant to them right now. The result was a meaningful improvement in how consistently users return to the app.

Smart Cost Controls

With clear visibility into what each part of the experience was costing, we redesigned the system to keep quality high while bringing running costs down significantly — and the savings have held as traffic has grown.

Built to Scale

The new platform was built with the next two years in mind. The Unlimits team can roll out changes to the AI behaviour with the same confidence as any other app update, with safety checks built in.

The Results

Significantly lower running costs — sustained as traffic grows, with room to scale further without costs growing in lockstep.

Improved user retention — driven by personalised, timely engagement instead of generic notifications.

Full visibility into cost and performance — the team can see exactly what it costs to serve any user, and resolve any issue, in real time.

A platform built to scale — designed for years of user history, with the safety and structure to grow without becoming fragile.

Technologies Used

LangGraph · n8n · OpenRouter · PostgreSQL · Redis · AWS