Xmartlabs | Our Work - CYTK: Building an AI-Powered Search Engine for the Automotive Industry

May 2020

Partnership begins

2022

Platform rebuild

Today

Ongoing collaboration

‍

Overview

Revolutionizing how repair information is accessed and used

CYTK has revolutionized how automotive repair information is accessed and used. Through a partnership with Xmartlabs that began in May 2020, CYTK transformed its initial voice-command app into a modern search platform that helps mechanics quickly find and understand the information they need.

From the start, the vision was bigger than search alone - CYTK is building toward becoming the central interface through which automotive technicians find, interpret, and act on repair intelligence - a mobile-first platform purpose-built for the shop floor.

CYTK brought deep domain knowledge of the automotive industry and a clear product vision; Xmartlabs brought the engineering depth and AI expertise to bring it to life. What followed was a holistic, collaborative rethink of how search and data architecture could serve industrial professionals at their best.

Challenge

Increasing efficiency in a rapid-moving industry

Automotive repair is a demanding and fast-paced industry that’s ever-evolving. There are vast amounts of technical information spread across multiple sources, and finding it is often time-consuming - manual searches through traditional repair manuals, complex terminology, and industry-specific knowledge slow down even the most experienced technicians.

CYTK aimed to speed up this process for automotive mechanics through technology. This came with a double challenge: not only did the platform have to be highly performant, but it also had to be intuitive for the diverse needs of automotive technicians, considering their unique working environment and preferences. Addressing this required a fundamental rethink of the platform’s architecture and AI strategy. Together, both teams aligned on three core principles:

‍

Search had to get smarter, not just faster

Mechanics search using industry shorthand, symptoms, and part names, not clean keyword queries. The platform needed to understand automotive language the way a senior technician does, returning the right result regardless of how the question was phrased.

Data architecture had to become the foundation

Heavy reliance on real-time calls to external data sources created unpredictable latency. The opportunity was to pre-process, standardize, and index data so that every query hit a purpose-built search layer, fast and reliable by design.

AI had to be applied where it actually adds value

The path forward wasn't more ML complexity, it was a clearer strategy for where intelligence belongs. Both teams aligned on a pragmatic principle: keep core search deterministic and fast, and apply AI precisely where it can meaningfully improve the user experience.

Solution

A holistic approach, built in close partnership

Working side by side as strategic consultants and technology partners, both teams rebuilt CYTK’s search capabilities across three interconnected pillars - each one informed by the domain expertise CYTK brought and the engineering rigor Xmartlabs delivered.

Rebuilt data pipelines & architecture

The first priority was the foundation. Together, the teams standardized data ingestion from multiple external APIs into a centralized taxonomy indexed on Elasticsearch, replacing fragile, real-time calls with a structured, pre-processed data layer that enables faster and more reliable queries. The result: a platform that doesn't just have the data, but can actually serve it at speed.

Optimized search for relevance & precision

With the data architecture in place, both teams turned to the search experience itself. They implemented synonym handling, intelligent scoring adjustments, pagination, and multi-source result merging, working through the nuances of how automotive professionals actually search, and building a system that meets them there. Search went from a pain point to a competitive advantage.

A targeted, pragmatic AI roadmap

Rather than adding AI complexity for its own sake, both teams defined exactly where intelligence creates real value. The guiding principle: keep the core deterministic, fast, reliable, and predictable, while applying Claude selectively where it meaningfully improves how users find information, understand it, and navigate complex workflows. For AI, they identified the use cases where Claude genuinely moves the needle:

Chat-based exploration

Mechanics ask questions in natural language and receive synthesized, context-aware answers from across the platform's data sources.

Manual & guide processing

Transforming complex repair documentation into proprietary, searchable content — making static materials part of the live search index.

‍

Keep the core deterministic — fast, reliable, predictable — while applying AI precisely where it meaningfully improves the user experience.

‍

04. Development accelerated with Claude Code

Today, that partnership continues at pace, now running on a purpose-built agentic workflow inside the codebase:

Role-based agents, not one generalist

A planner, a tester, a reviewer, and an orchestrator each work from their own slice of a modular CLAUDE.md, covering architecture, domains, coding standards, and CI.

Wired into the real workflow

MCP connections to Jira and Codegraph give agents live access to tickets and the codebase, not just static context.

Fewer approval loops

A Makefile and CI refactor let agents run trusted commands on their own, cutting the back-and-forth that used to slow the team down.

Memory that carries over

Learnings and open work persist across sessions, backed by living docs on testing and how concurrent agents coordinate.

‍

‍

Results

Record engagement, driven entirely by product quality

The rebuilt platform delivered immediate, measurable impact across every key metric. Notably, these resultswere achieved without active marketing campaigns; organic usage surged because the platform becamegenuinely faster, more accurate, and more useful.

27K+

Daily interactions

An all-time platform record, up 22% year-over-year

~91%

Monthly revenue retention

Across the active shop base

7.1

Vehicles loaded per shop

All-time record, up 27% year-over-year

~40K

Total vehicles

Indexed on the platform, also a record

‍

Deeper engagement, not just more searches

With improved taxonomy and relevance scoring, users weren't just querying more frequently — they were doing more with what they found. Average interactions per shop reached their highest level ever recorded.

A sticky, loyal user base

The improved experience translated directly into retention. Shops returned to the platform at 130% of the prior year's rate, with thousands of shops active the vast majority of their available months, a clear signal that the platform had become a trusted part of the daily workflow.

Leaner engineering, higher value

By aligning technology choices with actual business needs, the collaboration avoided unnecessary ML infrastructure and DevOps complexity. That same discipline now extends to how the team builds: with Claude Code part of the daily workflow, the lean engineering team sustains velocity across a growing roadmap without adding headcount.

Future-proofing for intent classification

Laying the groundwork for modern LLM-powered intent detection, to be activated once data maturity justifies it, building toward intelligence, not rushing into it.

‍

"We've had the same team alongside us for years, and their technical ability has always been exemplary. Now, working with Claude Code, that same lean team moves through our roadmap at a pace I didn't think was possible."

Bryan Levenson

Founder & CEO

CYTK

‍

Technology

Built for performance, designed to scale

Every technology choice was made deliberately - prioritizing reliability, maintainability, and fit for the demands of the automotive domain.

Search & Indexing

Elasticsearch with centralized taxonomy, synonym handling, multi-source merging, and intelligent scoring adjustments.

Data Pipelines

Python-based ETL pipelines ingesting and standardizing data from multiple external automotive APIs into a unified, queryable layer.

Infrastructure

Cloud infrastructure on AWS and Google Cloud, scalable, reliable, and built to grow with the platform's data volume and user base.

Mobile

Native iOS and Android applications that bring the full search experience directly to the vehicle, built for speed and reliability under real working conditions.

Development workflow: Claude Code

The team's current engineering workflow runs on Claude Code, accelerating day-to-day iteration across search logic, data pipelines, and mobile features, and sustaining the pace of a long-running roadmap with the same lean team CYTK has worked with since 2020.