I turn scattered business data into clean workbooks, useful reports and repeatable workflows. My services cover Excel cleanup and reconciliation, Python/VBA/SQL automation, dashboards and sales forecasting, website quality checks, and structured research.
As a former financial data analyst, I have extensive hands-on experience consolidating recurring reports, extracting database records, reconciling multiple sources, resolving historical data issues, maintaining formulas and delivering to strict reporting deadlines.
QA & TESTING: I have five years of QA experience, including website testing, Playwright and Cypress test automation, JavaScript/TypeScript, API testing with Postman, WCAG accessibility testing, Jira defect management and release acceptance testing. I focus on reproducible findings, clear test evidence and practical acceptance criteria.
EXCEL & DATA: Advanced formulas, lookup and conditional-summary functions, Power Query, PivotTables, dashboards, VBA, SQL queries and Python processing. I preserve identifiers, check totals and joins, document changes and separate uncertain records for review.
AUTOMATION & REPORTING: Reusable file-processing tools, report consolidation, database transformations and explainable business models, with clear instructions and validation of outputs.
WEBSITES & RESEARCH: Theme and interface improvements, Moodle consultation, reproducible QA findings, and source-backed software or industry comparisons.
You receive an agreed deliverable, clear written updates and a documented handover. Accuracy, confidentiality and deadlines are central to my work. My portfolio includes clearly labelled demonstrations with screenshots, methods and verification evidence; relevant working files can be provided for review.
Based in Canada. Chinese: native language. English: fluent.
Skills
Experience Level
Language
Work Experience
Education
Qualifications
Industry Experience
![Automotive research demonstration: comparison with traceable sources](https://www.twine.net/signin
Independent public-source research sample.
A product or sales team exploring automotive shop software needs a usable comparison, including the conditions behind feature and pricing claims. I prepared a focused research package covering Tekmetric, Shopmonkey and TireMaster, plus two relevant industry associations.
I organized official vendor and association information by workflow, deployment, inventory, integrations, pricing and channel relevance. Each claim in the evidence register has a source link, page location, review date, evidence status and follow-up question. Where a source was incomplete, I recorded an information gap instead of assuming the feature was absent.
The deliverables are a three-page English report and a CSV register containing 26 records across ten fields. The report connects the comparison to an illustrative repair-shop workflow and three commercial hypotheses, each with a practical validation step. The source snapshot is dated September 8, 2026.
Verification confirmed that all 26 register records matched the underlying research data and that the report’s 42 link annotations pointed to the nine registered official URLs. This gives a reader a direct route from a conclusion back to its basis.
The sample covers a selected set of products and channels. Vendor claims were not tested in live software, and the hypotheses have not been validated through customer interviews or market-size measurement.
The illustrated PDF and source register are prepared and available for review.
[Open the illustrated source comparison](https://www.twine.net/signin
![ERP validation demonstration: validate, confirm and prepare a draft](https://www.twine.net/signin
Synthetic offline demonstration.
Automated data entry needs a dependable decision process before information reaches an ERP. Missing suppliers, conflicting totals and repeated messages should be visible and reviewable.
I built a runnable prototype for purchase bills, supplier expense bills, structured sales commands and stock queries. It maps known master data, checks required fields and decimal totals, and places valid invoices in a confirmation queue. Confirmation is tied to the exact proposed content; changed previews and reused approvals are rejected. Duplicate checks persist across a local process restart.
The sample includes a two-page walkthrough, Python program, ten input scenarios, sixteen behavior tests and inspectable output files. In ordinary evaluation, three invoice proposals paused for confirmation and zero request drafts were exported. Scripted approvals then produced three local request candidates. The run also blocked one duplicate, routed five inputs to review and returned one clearly identified stock-fixture response. Replaying the showcase left the draft count unchanged.
All sixteen behavior tests passed in the recorded run. The exported requests remain unsent local candidates. Field names reference ERPNext version 15, but no ERPNext server, n8n instance, Telegram account or OCR service was connected. This demonstrates validation and confirmation logic; live integration and acceptance require the client’s configuration and environment.
The illustrated PDF, source code, input fixtures and validation records are prepared and available for review.
[Open the illustrated workflow evidence](https://www.twine.net/signin
![Sales forecast demonstration: scenarios, validation and refresh](https://www.twine.net/signin
Synthetic demonstration.
Sales planning needs a forecast that managers can inspect, update and compare with their targets. I created a working model that keeps sales volume, sales value and business assumptions connected.
Using 156 weeks of generated history across four product-and-segment series, I compared two seasonal methods on earlier validation windows, then froze the selected method before a separate 26-week holdout. This makes the method-selection process and its timing visible to the reader.
The deliverables are a six-sheet Excel workbook, a two-page case study and a command-driven weekly refresh tool with sample input and instructions. One scenario selector updates the same forecast model across pessimistic, realistic, stretch and optimistic cases. Product and segment totals reconcile to the business total, while actual weeks remain unchanged.
The recorded holdout unit WAPE was 5.34%, compared with 10.20% for the seasonal-naive benchmark. All four workbook scenarios reconciled to independent calculations. An added-week refresh preserved user assumptions, formulas and charts, and the refreshed file recalculated with zero formula errors in the tested spreadsheet engine.
These are demonstration results. Scenarios describe assumptions, not calibrated probabilities. The current model uses four fixed series and complete positive weekly data; desktop Excel and client-specific calendars still require acceptance testing.
The working workbook, refresh tool and illustrated PDF are prepared and available for review.
[Open the illustrated model walkthrough](https://www.twine.net/signin
![QA demonstration: checks, defects and evidence](https://www.twine.net/signin
Controlled demonstration with fictional data and deliberately seeded defects.
A software team needs more than a list of things that look wrong. It needs repeatable steps, clear impact and acceptance criteria that make fixes straightforward to verify.
I built and checked a small repair-shop interface covering appointments and work-order updates. I defined eight selected checks, recorded expected and observed behavior, and tested role-dependent screens and three application container widths. Each defect report connects the trigger to a screenshot, an explanation of impact and a proposed retest.
The finished sample includes a three-page QA report, a runnable local web fixture and a detailed execution log. The recorded run produced five passes and three failures: duplicate appointments after a repeated save, an advisor-only note visible to the simulated technician role, and a booking action clipped at 390 pixels. Successful checks include required-field validation and status persistence after reload.
The evidence was captured in one Windows desktop browser. Roles and widths are simulations, so the sample demonstrates workflow and layout QA rather than backend authorization or real-device coverage. The defects remain open; the retest criteria have been written, but fixes and post-fix retests have not been performed.
The illustrated PDF, execution log and runnable sample are prepared and available for review.
[Open the illustrated evidence page](https://www.twine.net/signin
![Shopify Liquid demonstration: typed specifications and real sample output](https://www.twine.net/signin
Technical catalogues need detailed product content that staff can maintain as new models are added. I built a reusable Shopify section that reads product metafield references and grouped metaobjects directly in Liquid. Typed specification rows preserve numeric 0 and boolean false, while missing values, empty groups and irrelevant product-type groups stay hidden. The deliverables include Liquid/CSS source, three synthetic product fixtures, portable HTML previews, 16 passing automated tests, installation instructions and a proposed migration/reconciliation plan for 40+ existing products. This is a self-initiated sample, rendered offline with LiquidJS. It has not been accepted in a live Shopify store or integrated with Electro 2.0; no client or commercial result is claimed.
Prepared materials: a two-page illustrated case study, Liquid section and snippet, CSS, synthetic fixtures, validation tests and installation notes. The PDF and working source package are available for review.
[Open the illustrated Liquid sample](https://www.twine.net/signin
Hire a Data Analyst
We have the best data analyst experts on Twine. Hire a data analyst in West Vancouver today.