Upwork / Portfolio Assets — Velat Isik¶
No prices appear here by design (brief rule: no unverified market prices in the repository). Pricing is positioning guidance, delivered separately.
Portfolio title¶
IngoShift — a source-traceable regional market-intelligence pipeline (DACH), with a quality gate that refuses to publish bad data
Portfolio description (approx. 850 characters)¶
Most research deliverables are spreadsheets that die on arrival. IngoShift is the alternative: a reproducible regional-intelligence pipeline for Ingolstadt/Region 10. It includes a verified live connection to the official TED procurement API, immutable raw archives, 35 quality rules, and a gate that refuses to publish when critical rules fail. Every metric can be traced to source bytes through content hashes. The polished bilingual dashboard adds English/German switching, filters, company profiles, tender/source details, and visible real/demo labels. It also includes conservative entity resolution, snapshot change detection, provenance-first Excel/CSV exports, and a reproducible weekly report. Built with Python and SQL, backed by 207 tests and strict typing. Synthetic examples are labelled everywhere; live data is never silently replaced by a fixture.
Longer case-study description¶
See case study — client-style problem, constraints, source governance, architecture, quality risks, and how the system adapts to a paying client's region and sources by configuration.
Service packages (deliverables only)¶
Package 1 — Data & Source Audit. Source inventory with per-source legal and access verification (evidence, not vendor claims); quality assessment of existing data; automation plan; sample normalized dataset; implementation estimate. Proof in repo: docs/source-catalogue.md discipline.
Package 2 — Automated Intelligence Pilot. One or two verified sources, normalized schema with lineage columns, validation rules with a publication gate, repeatable pipeline with one command, compact dashboard, handover documentation. Proof: the Phase 2–3 spine of this repo.
Package 3 — Full Monitoring System. Multiple sources, audited entity resolution, historized snapshots and change detection, dashboards, provenance- first exports, scheduled reports, tests, deployment, handover. Proof: modules 1–3 of this repo.
Paid pilot offer (structure)¶
Two weeks, fixed scope: one client source verified and integrated end to end — raw archive, validation rules, one mart, one export, one reproducible report — plus a written go/no-go recommendation for the full system. Client keeps the code and the source-audit evidence either way.
Proposal snippets¶
Data pipeline job: "You can audit any number I deliver: my pipelines store raw payloads content-addressed, stamp every record with source, parser version and data class, and a quality gate blocks publication on critical failures — in my portfolio repo that gate is proven by a test that poisons the data and watches publication stop."
Market research job: "I deliver the pipeline that produces, validates, updates, and explains the spreadsheet — rerunnable next quarter without me. The demo repo shows entity resolution with a full merge audit trail and a review queue instead of silent fuzzy matches."
Dashboard/BI job: "I build compact dashboards that people can inspect, not just admire: filters, record-level details, source links, retrieval dates, coverage notes, and visible quality state. The portfolio demo is a self-contained, responsive, bilingual file that can be hosted on GitHub Pages without a server."
Public-data integration job: "The portfolio includes a real official API integration, not only mock data: the TED collector stores the original response, parses the live multilingual schema, applies regional filters, and publishes only after the quality gate passes."
Client problems this project demonstrably solves¶
Research that cannot be rerun; numbers nobody can defend under audit; company lists full of duplicate/merged-wrong entities; "insights" built on silently missing data; reports whose method cannot be explained; vendor lock-in via opaque scoring (this system publishes components, never a black-box index).