Most Complex Build
🔗 Integration · Apex
WMS Bi-Directional Order Sync
The most technically demanding project of my career — syncing 10k+ orders/day between Salesforce and a legacy Warehouse Management System with zero data loss tolerance.
🔴 Challenge: The WMS had no idempotency support.
Duplicate inserts were silently corrupting order records and the client only found out during month-end
reconciliation.
🔵 Solution: Built a SHA-256 hash of every inbound
payload stored as a unique External ID. Any duplicate triggers an update instead of insert. Added a
Queueable retry chain with exponential backoff for transient failures.
🟢 Result: Zero duplicate orders in 14 months of
production. Processing time cut from 4 hours batch to under 90 seconds real-time.
Queueable Apex
Named Credentials
Platform Events
14 months production
📊 Measured Outcomes
90s
Order sync latency
Down from 4-hour batch window
30%
MQL lift via Pardot scoring
Calendly signal integration
60%
Manual data entry reduced
MISMO XML Batch pipeline
0
Deployment rollbacks
Post Copado CI/CD adoption
🗂 Key Project Stories
⚡ LWC
Inline Editing Data Table for Sales Ops
🔴 Challenge: Sales ops team was copy-pasting
between Salesforce and Excel to bulk-update 200+ records daily. Error-prone and eating 3 hours every
morning.
🔵 Solution: LWC inline-edit data table with
server-side pagination, SOQL wire adapters, and dirty-state tracking. Only changed fields sent to
server.
🟢 Result: 3-hour daily task down to 20 minutes. Zero
data errors reported in first 6 months of use.
🚀 DevOps
Zero-Regression Copado Pipeline
🔴 Challenge: 2–3 production hotfixes per sprint
due to manual deployments. Developers afraid to push changes on Fridays. No automated testing gate.
🔵 Solution: Implemented Copado with 4-environment
ring, mandatory PMD static analysis, and 75% test coverage gate before any promotion.
🟢 Result: Zero hotfixes in the 6 months following
adoption. Deployment confidence high enough that Friday releases became normal.
📈 Marketing
Pardot + Calendly Behaviour Scoring
🔴 Challenge: Static lead scores weren't reflecting
actual buying intent. Sales was ignoring MQLs because 60% were cold. Marketing and sales misaligned.
🔵 Solution: Webhooks from Calendly pushed booking
signals into Salesforce in real time. Pardot score boosted +80 on any booking, triggering immediate
sales alert.
🟢 Result: 30% MQL-to-pipeline lift in Q1 post-launch.
Sales started trusting the queue again — follow-up time dropped from 48h to 4h.
⚙️ Apex
MISMO 3.4 Mortgage XML Pipeline
🔴 Challenge: Loan processors manually re-keying
80+ fields from MISMO XML into Salesforce per application. Taking 45 min per file, 20 files/day.
🔵 Solution: Batch Apex that parsed deeply-nested
MISMO 3.4 XML using Dom.Document, mapped to custom objects, and surfaced validation errors in a review
UI.
🟢 Result: Processing time from 45 min to under 2 min
per file. 60% reduction in data entry errors in first month.
🛒 Commerce
Account-Based Pricing in B2B Commerce
🔴 Challenge: 500+ customers each with negotiated
price books. Standard B2B Commerce pricing didn't support the complexity. Storefront was showing wrong
prices for 30% of orders.
🔵 Solution: Custom Apex pricing extension that
resolves the correct price book per account at checkout time, with a cache layer to avoid SOQL on every
add-to-cart.
🟢 Result: Pricing errors dropped to zero. Page load
time on cart maintained under 1.2s despite the custom pricing logic.
⚡ LWC
Governor-Safe Bulk Flow Launcher
🔴 Challenge: Ops team needed to bulk-enrol 5,000
contacts into an onboarding Flow. Native solution hit CPU time limits at ~200 records and silently
failed the rest.
🔵 Solution: LWC with chunked processing (200
records/batch), Queueable chain for async execution, real-time progress bar via custom Platform Event,
and failed-record download.
🟢 Result: 5,000 contacts processed in under 8 minutes
with full audit trail. Zero governor limit errors since launch.
🔬 Deep-Dive: How Hard Problems Got Solved
🏭
🔗 Integration
The duplicate order problem
🔴 What went wrong: The WMS sent webhook retries on
any non-200 response. Salesforce occasionally returned 202 (accepted but processing) which the WMS
treated as failure and retried. We ended up with 3–4 copies of the same order record silently inserted.
🔵 How we fixed it: Changed the REST endpoint to always
return 200 immediately with a job ID, then process asynchronously. Added a SHA-256 composite hash
(orderId + timestamp bucket) stored as an External ID with upsert. Any duplicate payload now updates the
existing record instead of creating a new one.
🟢 What we learned: Never trust the upstream system to
handle idempotency. Build it into Salesforce regardless. The 20 minutes spent adding the hash field
saved weeks of data cleanup.
⚡
⚡ LWC
Why the data table felt laggy — and why it wasn't what we thought
🔴 What went wrong: After deploying the inline edit
table, users reported it felt "slow to respond" when editing a cell. We assumed it was the wire call. We
were wrong for long time.
🔵 The actual cause: The parent component was
re-rendering on every keypress because we'd put the dirty-state map in a reactive property. Every single
character typed triggered a full tree re-render. Fixed by moving dirty tracking into a plain
(non-reactive) JS Map and only flagging reactive state on blur.
🟢 What we learned: In LWC, reactive properties are
powerful but expensive. Keep internal component state in plain JS objects. Only promote to reactive when
you actually need the template to re-render.
📄
⚙️ Apex
The MISMO field that silently mapped to the wrong object for 3 weeks
🔴 What went wrong: One specific MISMO field — the
co-borrower's SSN last 4 digits — was being written to the primary borrower's record. The field path in
the XML was identical except for an ancestor node. Our XPath traversal was grabbing the first match and
stopping.
🔵 How we fixed it: Rewrote the parser to traverse the
full DOM tree and track parent node context before assigning values. Built a unit test suite with 12
edge-case XML fixtures covering co-borrowers, joint applications, and missing optional sections.
🟢 What we learned: For deeply nested XML, always
validate with real production samples — not the spec's simplified examples. The spec had 3 nesting
levels; production files had up to 9.
🚀
🚀 DevOps
The Copado deployment that passed all gates and still broke prod
🔴 What went wrong: A validation rule change passed
100% test coverage and PMD checks but broke a managed package's behaviour in production. Automated gates
didn't catch it because managed package tests run in their own namespace.
🔵 How we fixed it: Added a manual regression checklist
step in the Copado pipeline specifically for managed package touch-points. Also added a 30-minute
post-deploy monitoring window with Slack alerts on any new error logs.
🟢 What we learned: CI/CD gates are not a replacement for
domain knowledge. Automated checks catch what you've written tests for — human review catches everything
else.
Have a hard problem?
I work on complex Salesforce challenges — integrations, performance, architecture, and the bugs that have been "open for months". Let's talk about what you're building.