| π― Role | π¦ Domain | π Strength | π Location |
|---|---|---|---|
| QA Engineer | FinTech Β· AI Β· IDP Β· KYC | API + Data Validation | Pune, India |
|
Every defect I log has steps, expected vs actual, screenshots and source proof. Developers fix faster, nobody argues. |
I don't stop at the UI. I compare API response β UI β source PDF to catch the mismatches that slip through. |
OCR and LLM extraction can look right and still be wrong. I validate every field against ground truth. |
| π 100+ | π 50+ | π 30+ | π¦ 50β60 | β 120+ |
|---|---|---|---|---|
| Jira defects documented, tracked & retested | Test-case sheets from requirements & PRDs | Regression & change-validation cycles | Bank-statement formats across 4 markets | Manual test cases executed at UptoSkills |
flowchart LR
A[π Requirements / PRD] --> B[π§ Risk & Edge-case Analysis]
B --> C[π Test Case Design]
C --> D[βΆοΈ Execute: UI + API]
D --> E{Data matches source?}
E -- No --> F[π Log defect with evidence]
F --> G[π Retest & Regression]
E -- Yes --> G
G --> H[β
Sign-off]
style A fill:#1e3a8a,color:#fff
style F fill:#991b1b,color:#fff
style H fill:#166534,color:#fff
- π Manual, functional, regression, smoke, sanity & end-to-end testing
- π REST API testing with Postman, JSON validation & API chaining
- π OCR and LLM output validation for financial documents
- π¦ FinTech, Intelligent Document Processing & KYC/eKYC workflows
- π Python scripting for repetitive QA and test-data prep
- π Building automation skills with Playwright + TypeScript
Click to expand (format example)
| Field | Detail |
|---|---|
| Title | Closing balance extracted incorrectly for multi-page statement |
| Severity / Priority | High / P1 |
| Environment | Staging Β· Chrome latest |
| Steps | 1. Upload 3-page bank statement PDF β 2. Wait for extraction β 3. Compare closing balance with page 3 |
| Expected | Closing balance matches source PDF |
| Actual | Value taken from page 2 |
| Evidence | Screenshot + API response JSON + source PDF page |
| Impact | Wrong balance can affect loan decision |
pie showData title My QA Focus
"Manual & Functional Testing" : 40
"API Testing (Postman)" : 25
"OCR / LLM Data Validation" : 25
"Automation (learning)" : 10
- π Learning Playwright + TypeScript for UI & API automation
- π§ͺ Exploring how to test LLM / AI products (accuracy, hallucination, regression)
- π€ Open to QA Engineer / SDET-track roles in FinTech & AI