Leveret Systems
Technology Consultancy
AI agents · production-grade

We build the platforms that run your business

Custom software, system integrations, and data pipelines — built in weeks, not months. From replacing spreadsheets to full enterprise platforms.

UK-based Ltd companyRapid delivery cyclesArchitect-led team

What we do

Services

Each service below includes a live interactive demo — built from the same code and tools we use on real client projects.

Operations Control — Stock & Labour Balancing
LIVE
WIP value
£482k
2.4%
Labour util.
78%
1.1%
OTIF
94%
0.6%
Backlog hrs
330
3.2%
Stock by SKU · live
oklowreorder
0500100015002000100ST-01Steel236BR-12Bearings78MT-04Motors216FX-22M1013PN-07Hyd.64EL-30PLC
Labour utilisation · last 12 h
━ ━ cap 90%
0255075100-11h-8h-5h-2h-0h
Fabrication
55%
Welding
82%
Machining
100%
Assembly
70%
QA / Test
40%
Despatch
21%
WC-03 hits 96% capacity — reallocating 1 op from WC-06
BIM PIPELINE · PORTFOLIO

From source file to factory floor

Real models translated through Autodesk Platform Services and recorded straight from the SVF2 viewer. Fusion, Revit, IFC, point clouds — the same pipeline, drives our customers' Bills of Materials, work orders and digital twins.

Loading portfolio…

Who we are

The Team

We assemble the right team for every engagement — experienced practitioners across systems architecture, data analytics, software engineering, and digital marketing. Each project gets a dedicated team drawn from our network of specialists, led by a senior architect.

No bench warmers, no juniors learning on your project. Every team member is a practitioner with deep, hands-on experience in their discipline.

50+
Combined years of experience
4
Core disciplines
20+
Enterprise projects delivered
6+
Industry sectors

Systems Architecture & Development

20+ years experience

Enterprise platform design, full-stack development, system integration, and cloud architecture. Deep expertise in construction technology, manufacturing, and CAD-driven product lines.

Solution ArchitectureCAD/Product PlatformsBIM/Digital EngineeringPythonReactGCPNext.jsNode.js

Data Analytics & Business Intelligence

15+ years experience

Business analysis, BI development, data team leadership, and machine learning. Advanced qualifications including MSc Data Science. Track record at major enterprises in banking, telecoms, and fintech.

PythonSQLPower BIDAXMachine LearningData WarehousingBigQueryLooker

Software Engineering & DevOps

10+ years experience

Backend engineering, API design, CI/CD pipelines, and infrastructure automation. Experience across high-throughput platforms, real-time systems, and cloud-native microservices.

TypeScriptPythonDockerGCPREST APIsPostgreSQLRedisCI/CD

SEO & Digital Marketing

8+ years experience

Technical SEO, content strategy, and digital marketing across award-winning agencies. Sector expertise spanning e-commerce, travel, insurance, automotive, banking, and real estate.

SEO StrategyGoogle Analytics 4Looker StudioContent StrategySite MigrationsTechnical Audits

Selected work

Case Studies

UK Manufacturing CompanyBespoke product manufacturing · UK / USA / Canada / UAE6-week initial build · ongoing

From 15 spreadsheets to one BigQuery-backed control plane — built from their own data, in six weeks

440+ products · 20+ live projects · 16-discipline pipeline

A growing bespoke manufacturer was running its entire business — sales, engineering, purchasing, and the factory floor — across disconnected Google Sheets and an under-used CRM. Sales couldn't see which quotes were converting. Engineers re-typed Bill of Materials data from CAD into purchasing sheets. The factory's source of truth was a single shared spreadsheet covering 16+ trade disciplines and 20+ live projects across four countries. Parts shortages were discovered the day before despatch.

6 wk
Initial build cycle
440+
Products in catalogue
60k+
Emails AI-classified
20+
Live projects · 4 countries
16
Discipline gates per WO
78+
API endpoints built
0
Manual Bill of Materials transcription
1
Source of truth (was 15+)
system_architecture.erd · BigQuery + Cloud Run
LIVE
Entities
16
Relationships
33
Cloud Run
3
Throughput
1.3k ev/s
EXTERNAL · SOURCESAPPLICATIONS · COMPUTESALES INTELLIGENCE · BIGQUERYCAD CATALOG · BIGQUERYOLTP & MANUFACTURING · CLOUD SQL + BIGQUERYSTORAGE · GCSEXTERNALWebsitequote-builderGA4 eventsEXTERNALInboxGmail · sales@60k+ historyEXTERNALCRMopportunitiesevents logEXTERNALXeroinvoices · paidEXTERNALCAD packspack-and-goAPP · CLOUD RUNSales Dashboardeurope-west1Flask · ~78 APIsAPP · CLOUD RUNCatalog Dashboardeurope-west2Drafts · APSAPP · PWAManufacturing PWA/mfg/ · Cloud RunOffline · 12 screensVM · COMPUTECAD Worker VMWindows · port 8081Pack-and-Go · APS uploadTABLEdg_quotesPK quote_idga_client_idgrand_totalTABLEdg_crmPK opp_idstage · probabilityexpected_closeAI · TABLEdg_emailFK contact_idintent · sentimentai-suggested draftTABLEdg_xeroPK invoice_idpaid / dueEXPORTga4_*events_YYYYMMDDuser_pseudo_idVIEWdg_analyticscustomer_journey · pipeline_forecast · funnel_metricsTABLEmodel_filesPK model_idaps_urn · viewer_urldrive_folder_idTABLEmodel_componentsFK model_iddepth · materialteam · supplierTABLEmodel_bomFK model_idqty per base_nameBill of MaterialsTABLEparts_libraryPK normalizedconfidence scorecategoryTABLEcatalog_productsPK variation_skuteam flags · archivedTABLEproduct_documentsrenders · plansspec sheets PDFTABLEprice_listsGBP / USD / CADFK variation_skuOLTP · CLOUD SQLPostgreslive work-order stateuser sessions · authTABLEmfg_projectsPK project_idcountry · valueTABLEmfg_work_ordersFK project · sku8-gate stateTABLEmfg_progressFK wo_idphotos · hoursTABLEmfg_usersrole · emailteamGCSmfg-photospublic · per-WOGCSmodel-metadataaudit JSONGCSmodel-zipsAPS input ZIPsGCSwebsite-assetsrenders · spec PDFsEXTAPSSVF2
SalesCatalogManufacturingPostgres OLTPStorageCompute / ServiceExternalsolid: direct write · dashed: scheduled sync

Before · the pain

  • Sales pipeline split across the inbox, a website quote-builder, the CRM, and Google Sheets — no unified view of any client
  • 60,000+ historical emails containing orders, objections, and missed follow-ups that nobody could systematically read
  • Bill of Materials data manually transcribed from CAD into purchasing sheets — typos caused parts shortages mid-build
  • A single shared planning spreadsheet was the entire factory's source of truth: 16 trade disciplines, 20+ projects, no real-time view
  • Parts coverage tracked in a separate workbook; shortages discovered days before despatch deadlines

After · centralised on GCP

  • One BigQuery warehouse stitching GA4 sessions, website quotes, inbox, CRM, Xero, and CAD — every business surface joined on the customer record
  • An AI agent (Claude Haiku + Sonnet) classifies every incoming email — intent, entities, sentiment, suggested reply — and surfaces missed opportunities in the dashboard
  • Bill of Materials data auto-extracted from CAD via Autodesk APS, normalised in the parts library, and pushed to the factory app — engineers stop re-typing
  • A factory-floor PWA replaces the planning spreadsheet: real-time work orders, 8-gate pipeline, daily photo progress, offline queue when the wifi drops
  • A Coverage Tree across 20+ live projects shows every part's allocation, what's on PO, and what's short — surfaced before it becomes a delivery risk

Systems built · what powers the transformation

all on GCP · OAuth-gated · BigQuery-backed

Sales Dashboard

live
Cloud Run · europe-west1

Flask + BigQuery sales-intelligence workbench behind Google OAuth. Five surfaces: What's Working, Pitfalls, Client Journeys, Planning, Catalog. ~78 API endpoints, Xero actuals overlay, probability-weighted pipeline forecast.

GA4CRM syncEmail AIXero78 APIs

Catalog Dashboard

live
Cloud Run · europe-west2

Product-master workbench: AI-generated descriptions, 3D model extraction via Autodesk APS, normalised parts library, draft/approve/publish workflow, auto-generated compliance spec sheets.

CAD → APSAI specsDrafts440+ products

Manufacturing PWA

live
/mfg/ · same Cloud Run · catalog-driven

Factory-floor PWA replacing the planning spreadsheet. Projects, work orders, 8-gate discipline pipeline, daily photo progress, offline queue (IndexedDB), role-based UI (worker / lead / admin), live SKU picker auto-fills the Bill of Materials.

Offline-first12 screensPhoto progressSheet writeback

GCP Substrate

live
Two projects · three Cloud Run services · one VM

Everything centralised: BigQuery warehouses, GCS buckets for photos / metadata / spec sheets, a Windows VM running the CAD Document Manager for assembly pack-and-go, Autodesk APS for SVF2 translation. Deploy from master, OAuth-gated, secrets in env.yaml.

BigQueryCloud RunGCSAPSCloud Scheduler

For the factory floor

Automation does the repetitive, error-prone work

  • Operatives open the PWA on a tablet — see today's work, capture progress photos, update the gate, log hours; works offline and syncs when reconnected
  • SKU picker auto-fills the work order name, documents, and Bill of Materials from the published catalogue — no manual re-entry
  • Daily photos are publicly URL-addressable in GCS; the daily progress banner shows the day's outputs across the team
  • 8-gate pipeline (DSGN → GARV → MTRL → LASR → … → SHIP) is visible per work order — everyone knows what's blocking what

For management

BI you can actually act on — every system unified

  • Live Coverage Tree: every part across every project, % allocated, unallocated £, on PO, days-to-deadline — one click to drill into any project
  • Pipeline forecast with probability weighting, Xero actuals overlay, cycle-time analytics, win/loss heatmap by product and country
  • Missed-opportunity report from the AI email agent — quotes with no follow-up, deals presumed lost, recovery scenarios
  • Per-project work-order progress with photo evidence, hours booked vs estimate, materials status — a director can audit any project in 60 seconds

Deliverables

Sales Intelligence dashboard with 5 analytical surfaces and ~78 API endpoints
Manufacturing Manager PWA replacing the central planning spreadsheet (offline-capable, 12 screens, role-based)
CAD → APS → BigQuery extraction pipeline with auto-generated compliance spec sheets
AI email agent: Claude Haiku triage + Sonnet extraction on 60,000+ historical emails
Coverage Tree across 20+ live projects with parts allocation, PO tracking, and shortage alerts
Xero hourly sync via Cloud Scheduler — actuals overlay on pipeline forecast
Centralised on GCP: 2 projects, 3 Cloud Run services, BigQuery warehouses, GCS buckets, Windows VM, Autodesk APS

AI in production

Six AI agents shipped — running every day, in production

Not demos. Not chatbots. Real production agents reading our clients' data and writing actions back into their stack — email inboxes, Revit models, supplier feeds, customer journeys, internal PRs. Built with model routing, retries, observability, and hallucination guards. Two-pass pipelines (Haiku triage → Sonnet depth) keep cost down without losing accuracy.

Requests · 24h
4,280
Tokens · 24h
142M
Median latency
640ms
Cost saved
£8.4k / month

Email Triage Agent

LIVE

Two-pass classification on every inbound — intent, sentiment, entities, draft reply.

Inbound emailHaiku triageSonnet extract→ CRM record + draft
HaikuSonnetBigQuery
60,124 emails · 99.1% triage accuracy

Revit Element Classifier

LIVE

Native Revit properties + position rules → 38 derived element classes.

Federated modelPosition rulesSonnet review→ classified BigQuery view
SonnetForge SDKRules engine
4,200+ elements · 38 derived classes

Supplier Risk Forecaster

LIVE

Predicts slip risk per supplier from historical lead-time variance + live signals.

Supplier feedsML modelSonnet narrative→ RAG dashboard
XGBoostSonnetSupplier feeds
12 feeds · daily forecasts

Customer Journey Stitcher

LIVE

Unifies every GA4 session, quote, email and CRM event per customer — semantic match.

Multi-source eventsEmbeddingsRAG retrieval→ journey timeline
EmbeddingsVector DBSonnet RAG
880k events stitched

Quote Draft Assistant

LIVE

Turns a customer enquiry into a draft quote with line items priced from catalogue.

Customer briefCatalogue lookupSonnet drafts→ Editable quote
SonnetCatalogueTool-use
Drafts in 30s vs 25 min

Internal Dev Copilot

BETA

Reviews every internal PR against codebase conventions + flags risky migrations.

PR openedRepo contextSonnet review→ PR comments
SonnetRepo indexPR comments
Median PR cycle −38%
All agents observable in production: tokens / cost / latency / accuracy / hallucination guard triggers.

AI engineering

How we keep AI consistent in production

LLMs are extraordinary at reasoning, synthesis, and finding needles in haystacks of unstructured data — but on their own they're inconsistent and burn tokens. We build them inside a three-layer hybrid stack: deterministic rules narrow the haystack, ML ranks the candidates, and the LLM does the novel reasoning at the end — always with citations, schema-validated outputs, and full observability.

hybrid_stack.runtime · needle through the haystack
SAMPLE QUERY · 1 / 3
USER QUERY
"Which suppliers risk slipping this month?"
HAYSTACK
Raw data — your entire warehouse
BigQuery · Postgres · GCS · embeddings index
10.4 M rows
all supplier events across all projects, all time
L1 · DETERMINISTIC
Cheap, predictable, consistent
SQL filters · rules engine · validators · scope guards
10,432 rows
SQL: active projects · current month · scope filter
L2 · ML / STATISTICS
Pattern recognition & ranking
XGBoost · Random Forest · embeddings · vector search · classifiers
47 candidates
XGBoost slip-probability + embedding similarity
L3 · LLM REASONING
Synthesis · novel reasoning · explanation
Claude Sonnet · function calling · RAG over the candidates · cited output
determinism narrows · ML ranks · LLM reasons + citesschema-validated · token-traced · hallucination-guarded

Architectural principles

01Hybrid: rules + ML + LLM

Determinism where it matters (PK lookups, validation, scope filters). ML for forecasting and ranking. LLM only for novel reasoning and synthesis. The right tool per task — never a hammer for every nail.

02RAG with citations

Vector search over your own data + LLM synthesis. Every claim cites the source row, document, or chunk. The model can't fabricate — it can only retrieve and combine. If the data isn't there, the agent says so.

03Two-pass cost optimisation

Haiku triages cheap and fast — 80% of cases never touch Sonnet. The deep model only sees what genuinely needs reasoning. ~4× cheaper without losing accuracy. Per-inbound cost: pennies.

04Structured outputs · function calling

JSON schemas enforced at the model boundary. Function calling so the model returns tool invocations, not prose. Downstream code never parses free-text — fewer surprises, easier rollback.

05Hallucination guards

Output validated against schema. Citation requirement on every claim. Confidence thresholds — low-confidence outputs route for human review, never auto-commit. Drift detection on output distributions.

06AI as the orchestrator, not the worker

The LLM chooses which rule to apply, which ML model to consult, which dataset to query. It steers the stack and reasons over outputs — it doesn't replace the deterministic plumbing underneath. Reasoning > replacement.

Engagement

Pricing

Flexible engagement options to match your project scope and budget.

Day rates

Architect & LeadArchitecture, design, build, advisory
£800–950/day
BI / Analytics SpecialistData-driven projects, reporting, ML
£350–450/day
SEO / Digital MarketingWebsite builds, digital marketing
£300–400/day
Full-stack DeveloperPlatform engineering, API development
£400–600/day
Fixed-price deliveryWell-defined projects
Scoped per project

Indicative project pricing

Systems audit & architecture roadmap
1–2 weeks£5,000–£8,000
Single-platform build or integration
4–8 weeks£15,000–£30,000
Enterprise platform programme
3–6 months£50,000–£120,000
AI/ML proof of concept
2–4 weeks£10,000–£25,000
Website build (React/custom)
2–5 weeks£8,000–£20,000

Industries

Sectors

Construction & AEC

BIM/CDE integration, ISO 19650, Autodesk APS, digital engineering

Manufacturing

Production planning, quoting systems, inventory and pricing management

Utilities & Infrastructure

Asset management, OT/IT integration, digital twin platforms

Professional Services

Client portals, document automation, workflow management, CRM

SMEs & Scale-ups

Replacing spreadsheets and manual processes with custom platforms

Process

How we work

01

Discovery call

We understand your problem, systems landscape, and goals.

02

Architecture & scope

A clear proposal with architecture, timeline, and fixed or day-rate pricing.

03

Build

Iterative delivery with weekly demos — you see progress from week one.

04

Launch & support

Deployment, training, documentation, and ongoing support options.

Get in touch

Start a conversation

Whether you have a specific project in mind or want to explore how automation could help your business, we would love to hear from you.