Arindam Chatterjee

Self-initiated web application AI-assisted personal project, not a client commission. Every project, client name, response and award shown in the screenshots is synthetic. No live office records or client data were used. Contribution and stack The interface was developed in my workspace through requirements-led iteration with GitHub Copilot-assisted coding and testing. The implementation uses semantic HTML, CSS, vanilla JavaScript modules, Lucide icons and Manrope typography. Node.js and esbuild support the local toolchain. No external human collaborator delivery or commissioned agency website is claimed. What is demonstrated Clickable summary metrics filter the correct records. Search and client-status filters update the list. Detail and edit dialogs expose scope and evidence. An award transition is rejected without agreement evidence. Desktop tables and compact mobile views present the same workflow. Verified checks A reproducible Playwright runner exercises the real frontend with an in-memory synthetic API using the existing validation model. It verifies metric navigation, editing and reload, search, required-evidence rejection, working dialogs and zero horizontal overflow at 1440, 390 and 320 pixels. The run finished with zero page errors and zero external requests. Screenshots and their SHA-256 hashes are retained with a verification receipt. Limits and delivery approach The screenshots are a frontend demonstration, not a production backend, CMS, real-payment or external GitHub-sync test. No paid-client outcomes are claimed. For client work I would first agree the design, content, target stack, accessibility and acceptance checks, then deliver a bounded implementation with tests and handover notes. AI assistance and handling of client material require prior written agreement.

• Software Engineer

Self-initiated demonstration AI-assisted software example, not a client commission. All input documents and values are synthetic. What it does Three generated AcroForm PDFs are mapped into one XLSX workbook using an explicit six-field schema. The implementation uses Node.js, pdf-lib and ExcelJS. It preserves leading-zero identifiers as text, numeric amounts, boolean checkboxes, blank values and multi-select arrays. Formula-looking and XML-like text is retained literally. Evidence The saved workbook was read back and checked against the expected values and types. Input and workbook SHA-256 hashes and a JSON verification receipt are retained. Four extraction regression tests passed. The image shows the actual synthetic workbook readback, not invented client results. The sample was tested locally on macOS. Limits No OCR, scanned/flattened/encrypted forms, XFA or high-volume production certification is claimed. A real project needs authorized representative files, an agreed field map, acceptance checks, scope, price and payment terms before work. Prepared with GitHub Copilot assistance and checked with automated tests and visual review. No external client or collaborator delivery is claimed.

• Software Engineer