From Hours of Manual Search to Verified Answers in Seconds

How AIM Deployed a Production-Grade AI Knowledge Assistant for Mirlin Fleet Management — Grounded in Their Own Documentation, Live in Production
Join Beta Now

Executive Summary

Mirlin Fleet Management's workforce — dispatchers, technicians, and drivers — navigated over 1,000 pages of fleet management documentation every day. Finding a single procedure, specification, or policy meant hours of manual search, inconsistent answers across employees, and field workers stopping work to call the office for help.
AIM built and deployed an enterprise AI Knowledge Assistant purpose-built on Mirlin's own documentation. Every question now receives a verified, source-cited answer in seconds — available via text or voice, on any device. Answers are grounded exclusively in Mirlin's documents: zero hallucinations, 100% traceability, live in production.
This case study covers the specific problem Mirlin faced, how the solution was designed and deployed, and the measurable impact for every level of the organisation.
AIM SmartAssist Widget
tell me about salvation army
12:55 PM
Shift+Enter for new line
Knowledge Base
Manage and index your document library
Processing through AI pipeline...
https://donate.salvationarmy.ca/page/63606/donate/?_ga=2.70286800.139...
Cancel
0 of 50 pages indexed 0m
No documents yet. Upload files to build your knowledge base.

The Challenge

The problem wasn't a lack of documentation. It was a lack of access.

Mirlin's operational knowledge existed — captured across hundreds of fleet procedures, maintenance specifications, compliance policies, and operational tables. But no one could reliably find what they needed, when they needed it.

Three compounding problems

Knowledge Sharing

Knowledge was buried, not accessible.

Over 1,000 pages of fleet documentation spanned multiple formats and repositories. A technician on a job site needing a torque specification had no fast path to the answer. A dispatcher fielding a compliance question had to dig through binders or call a subject matter expert — if one was available.

Answers

Answers were inconsistent — and unverifiable.

Different employees produced different answers to the same question, depending on who was asked and which document they found. There was no single source of truth, no mechanism to verify that guidance matched documented policy, and no audit trail for compliance-sensitive queries.

High Risk

Experienced knowledge was at risk.

Critical operational knowledge lived in the heads of long-tenure employees. When those employees left, the knowledge left with them. There was no way to capture, structure, and make that knowledge reliably accessible to the whole team.

For field workers, the impact was most acute. A technician mid-job needing a specification had two options: make their best guess or stop everything and call the office. Neither was acceptable from a safety, quality, or efficiency standpoint. The documentation existed — but it might as well not have, for anyone in the field without time to search through it.

The cumulative cost: hours of lost productivity per employee per week, elevated operational and compliance risk, and a growing gap between what Mirlin's documentation contained and what the workforce could actually access and use.

The Solution

An AI Knowledge Assistant built on Mirlin's own documentation

AIM deployed an enterprise AI Knowledge Assistant that fundamentally changed how Mirlin's workforce accesses operational knowledge. The system answers questions in plain language, retrieves answers exclusively from Mirlin's uploaded documentation, and cites the exact source for every response.

Three design principles that make it work

1st

Grounded intelligence — zero hallucinations by design.

The system does not draw on general internet knowledge. Every answer comes exclusively from Mirlin's uploaded documents. This is not a configuration setting — it is an architectural constraint that makes hallucinations structurally impossible. Every answer is sourced. Every source is cited.

Circled 2

Dual-store retrieval — precise answers from mixed content.

Fleet documentation contains both structured data (tables of specifications, parts lists) and unstructured content (narrative procedures, policies). AIM's dual-store architecture routes each query type to the appropriate retrieval pipeline — structured data to one store, narrative content to another — then ranks and combines results for the final response.

Circled 3

Voice-first design for field access.

Office workers ask questions via text. Field technicians and drivers ask via voice — hands-free, without stopping work. The system processes spoken queries, retrieves the relevant answer, and responds in natural language. Multi-platform delivery (web portal, iOS, Android) ensures access from any context, on any device.

How the architecture is organised

  • User Interaction Text chat, voice input and output, web portal, mobile apps (iOS/Android)
  • Knowledge Assistant Query analysis, context tracking, citation extraction, response orchestration
  • AI Services Language model (grounded to organisational documents), content safety, semantic search
  • Knowledge Systems Dual-store retrieval pipelines (structured + unstructured), semantic index, content cache
  • Enterprise Integrations Document repository (PDF, DOCX, XLSX, PPTX, HTML), file storage, org systems
Security & GovernanceData isolation per org, encryption in transit and at rest, role-based access, audit logging, responsible AI by design — zero hallucination grounding, 100% source traceability, multi-tenant data isolation, role-based access, and content safety built into the architecture

Who Benefits

Who BEvery layer of the organisation gains something specificenefits

Role

Field Technicians & Drivers

Before

Stop work → call office → wait → hope the answer is right

After

Voice query → verified answer in seconds, hands-free

Dispatchers & Operations

Manual search through hundreds of pages

Instant, consistent procedure answers, 24/7

HR & Compliance Teams

Repetitive questions, no audit trail, no source verification

Self-service answers with 100% traceable citations

Training & Onboarding

New employees waiting for expert guidance

On-demand access to the full knowledge base from day one

Leadership & Executives

Knowledge locked in long-tenure staff; lost when they leave

Institutionalised knowledge — accessible, auditable, permanent

How It Was Deployed

From zero to production through five structured phases

Audit

Knowledge Audit

Assessment of Mirlin's full documentation landscape: 1,000+ pages across PDF, DOCX, XLSX, and PPTX formats. Content type analysis (structured tabular vs. unstructured narrative). Stakeholder mapping and definition of query taxonomy and acceptance criteria.

Knowledge Sharing

Knowledge Architecture & Ingestion

Document preparation (text-based PDFs, proper heading structure, clean table formatting). Separate ingestion pipelines configured for structured and unstructured content. Semantic indexing and retrieval validation against representative query sets.

AI

AI Configuration & Governance

Language model grounding — responses restricted to organisational documentation only. Content safety and prompt handling policies applied. Source citation architecture implemented. Responsible AI properties confirmed end-to-end: grounding constraint (zero hallucination), 100% source traceability, data isolation, role-based access, and content safety — governance embedded in the architecture.

Merge Documents

Interface & Voice Integration

Employee web portal and admin portal deployed. Voice input/output integrated for field workers. Multi-platform rollout across desktop, iOS, and Android. Role-based access configured per stakeholder group.

Youtube Live

Production Validation & Go-Live

User acceptance testing across all stakeholder groups. Zero-hallucination and source traceability validation. Sub-second response time confirmation. Go-live with change management and onboarding support.

Results

Three production metrics. One clear outcome: it works.

    SECONDS

    Response Time

    Questions that previously required hours of manual search now receive verified answers in seconds

    100%

    Source Traceability

    Every AI-generated answer cites the exact document and section — no answer is unverifiable

    ZERO

    Hallucinations

    All responses are grounded exclusively in Mirlin's documentation — structurally impossible to fabricate

    LIVE

    In Production

    Not a prototype. Not a pilot. A production system actively used across the organisation

    Operational gains across the workforce

    • Field technicians and drivers access fleet knowledge hands-free via voice while on the job
    • Dispatchers and operations teams receive consistent, verified answers — not individual interpretation
    • HR and compliance teams have a self-service knowledge layer with a full audit trail
    • Onboarding time reduced: new employees access the complete knowledge base from day one
    • Institutional knowledge preserved: critical operational expertise is no longer locked in individuals

    The productivity shift is not incremental — it is structural. When knowledge retrieval moves from hours to seconds, the economics of every knowledge-intensive workflow change. Employees stop waiting. Teams stop second-guessing. Compliance stops being reactive.

    Strategic impact for leadership

    Operational resilience

    Mirlin's knowledge base is now institutionalised — accessible regardless of workforce turnover

    Compliance readiness

    100% source traceability means every answer can be audited and verified

    Productivity at scale

    The time economics of knowledge access transform when answers take seconds instead of hours

    Production-grade responsible AI

    Grounding, source traceability, data isolation, and access controls are architectural properties — demonstrable, not just described

    Why AIM

    60% of generative AI proofs of concept never reach production.

    The gap between a working prototype and a production system is where most enterprise AI projects fail — hidden barriers in governance, security, organisational readiness, and deployment discipline.

    AIM's approach is production-first from day one. The Mirlin Knowledge Assistant was not built as a demo. It was built as an enterprise system:

    • Responsible AI built into the architecture: zero hallucination grounding, source traceability, multi-tenant data isolation, role-based access, content safety, and full audit trail are structural properties — not bolt-on policies that can drift or be bypassed
    • Text chat, voice inZero-hallucination by design: Grounding is an architectural constraint, not a configuration that can driftput and output, web portal, mobile apps (iOS/Android)
    • Certified enterprise AI practice: Anthropic Governance Alignment | Botpress Certified Partner | OpenAI Frontier Expertise
    • 19 years of enterprise delivery: Architecture In Motion Inc. brings a track record of deploying production systems in complex enterprise environments

    Most enterprise AI engagements fail not because the technology doesn't work — but because deployment discipline, governance readiness, and change management are treated as secondary concerns. AIM treats them as primary. Every deployment begins with a governance framework in place, a defined acceptance criteria for production readiness, and a structured path from pilot to live system.

    The Mirlin Knowledge Assistant is live. It is in use. It delivers results that can be measured. That is the standard AIM holds every engagement to.

    Ready to Transform How Your Workforce Accesses Knowledge?

    Is your organisation sitting on valuable knowledge that your workforce can't easily access?

    If your teams spend hours searching for answers that exist somewhere in your documentation — or if you've seen AI pilots that never made it to production — AIM can help.

    We specialise in enterprise AI deployments that go live, deliver measurable results, and meet the governance, security, and reliability standards your organisation demands.

    Every engagement starts with a clear-eyed assessment of your knowledge landscape, your workforce workflows, and your governance requirements. We bring the architecture, the deployment discipline, and the production track record. You keep full ownership of your data — the AI never reaches beyond your documentation.