I'm Jorhan, a data analyst who turns messy data into clear stories.

Earned-media exports arrive with scrambled or missing city and state data, so local market credit can’t be trusted.

The code verifies the source of every media hit using its call sign and URL, not the pre-filled city and state. The state is the double check, so a “Cleveland, AL” never lands in Cleveland-Akron. The dataset needs to be local only, so national and wire outlets are flagged instead of forced into a market, which keeps them from contaminating it. Every decision is logged in an audit column: four overlapping checks in all.

A reusable Claude Project with Nielsen-standard formatting and color-coded change tracking. On a single 10,000-row export, 99.3% of rows came back accurate on the first pass in under 5 minutes, work that would have taken 7+ hours by hand.

Approximately 15 hours per week of time saved for the measurement team.

Python, data QA, Claude instruction design

Choosing which past campaigns to benchmark against is judgment-heavy and leans on whoever remembers the history.

A five-step workflow: upload the intake form, gather context, search the historical database, propose 2 to 3 candidates with reasoning, then write goal ranges back into the form. Guardrails: verify in Meltwater, flag campaigns older than about a year, never invent data.

A completed Claude Project that gives someone with no campaign history strong benchmark candidates, grounded in the historical database.

Saves approximately 2 to 3 hours per campaign and speeds up benchmark decisions by about 50%.

Process design, benchmarking methodology, AI workflow design

Daily Meltwater exports needed consistent classification across dozens of products and three campaign phases.

A nine-step pipeline: parse, exclude, dedupe syndication, classify coverage type, assign vertical, detect syndication origin, match against the 272-outlet target list, score products, route to the right phase tab.

A stable daily process feeding a structured Excel tracker for pre-Prime, Prime and post-Prime.

At least 6 hours per day saved during the peak week.

Python, ETL design, media classification

Broadcast volume was inflated by paid ads, commercials and unrelated mentions.

A reusable Python classifier plus a matching Boolean exclusion query.

First major run on 3,263 rows: 2,630 removed, 171 held for review, 462 kept. Totals lined up with the revised KPI goals, and a full categorization workbook serves as the audit trail.

Now used across campaigns, it speeds up building new Meltwater dashboards.

Data cleaning, Boolean logic, QA

Brands need to know whether they own the answer in AI platforms, not just in search or news.

Tracked campaigns across all four major LLMs and benchmarked them against a set of 40 historical campaigns, to see how a particular campaign shows up in AI and LLM search.

Short, stakeholder-ready summaries that stay honest about what the data supports, including where a superlative claim wasn’t backed up.

AEO, benchmarking, narrative analysis

Multiple concurrent campaigns, tight deadlines and shared monitoring duties.

An automated daily Hot Sheet: an interactive HTML coordination tool with color-coded assignments and print support.

A clean email-optimized version goes out to the team, so everyone sees the same picture each day.

Project coordination, HTML tooling, stakeholder communication

Updating campaign results by hand was slow and error-prone.

Built the database and an import workflow, so a completed report’s numbers load straight in.

A single source of benchmark data that also powers the benchmarking assistant.

Approximately 5+ hours per week saved.

Data modeling, Excel, process automation

80+ sites, possibly competing for the same searches.

Keyword gap analysis across every domain.

The sites were cannibalizing each other. After consolidating, traffic, conversions and bounce rate all rose 43% in a test, and the change was rolled into the main site.

Switch listening platforms without breaking national campaign reporting.

Rebuilt the setup, audited data discrepancies, kept the client informed.

Reporting continued with no disruption.