Spark Analytics · Baguio City, Philippines
I build the pipeline behind your dashboard.
I set up scheduled data collection, clean the data in SQL, and report it in Power BI or Excel. The numbers refresh on their own, and an hourly check tells you when a job stops.
What I do
Scheduled collection
Data pulled from APIs and public pages on a timetable: YouTube Data API, Instagram Graph API, TikTok, Spotify, Deezer. Jobs run on pg_cron, GitHub Actions or a small Linux server, whichever fits the source.
A clean SQL model
Raw tables keep the data exactly as it arrived. Staging views clean it, and mart views answer the business questions. PostgreSQL on Supabase, with every step in versioned SQL files.
Dashboards people can read
Power BI, Excel (pivots, slicers, VBA) or Tableau Public. Labels in plain words, and the hover text on each chart says what the number means and where it comes from.
Health checks
Every job is checked each hour and marked OK, Late or Down, so a broken feed shows up the same day instead of in next month's report.
Client accounts
Connecting a client's social accounts, with consent
For client work, Spark Analytics reads a brand's own TikTok, YouTube, Instagram and Facebook numbers through each platform's official API. Nothing is connected without the account owner's written permission and their own click on the platform's consent screen.
The owner clicks Allow
The account owner signs in on the platform's own page and approves the access. Spark Analytics never sees or stores a password.
Read-only
Followers, posts and their view, like, comment and share counts. No posting, no messaging, no ads, no changes to the account.
Private by default
Numbers are stored once a day in a private database and shown only to the account owner and their team. Never sold or shared.
Revoke any time
The owner can remove access from the platform's settings, or ask for the data to be deleted.
Why the API and not the native apps: the native analytics screens show one account at a time, keep limited history and cannot be combined across platforms or reported on automatically. Details are in the privacy policy.
Case study
Social analytics for an OPM music channel
A daily pipeline that tracks a live-session music channel and the artists it could invite next. YouTube, Instagram and TikTok numbers arrive every morning; Spotify, Deezer and kworb charts every week. Five benchmark channels (Wish 107.5, Backyard Live, THE FIRST TAKE, it's Live, NPR Tiny Desk) give the numbers something to be compared with.
- SourcesYouTube · Instagram · TikTok · Spotify · Deezer · kworb
- Schedulerspg_cron · GitHub Actions · Home server cron
- RawData as it arrived · Never edited
- StagingCleaned · Typed · De-duplicated
- MartOne view per question
- DashboardPower BI · Plain-word labels
What it answers
- Which posts beat the channel's usual numbers, on each platform, and by how much.
- Which artists fans ask for in the comments but who have never played the channel.
- Which OPM artists to invite next, ranked by how often their songs are used on TikTok, their Spotify charts and their follower growth.
- How the channel compares with the benchmark channels on views per upload and posting rhythm.
Days before collection started are filled with labeled synthetic rows, kept in their own schema so they never mix with real data. Independent analysis of public data, not affiliated with Nine Degrees North.
How a project runs
Questions first
We list what the dashboard has to answer before any code is written.
Find a source for each
Every question gets a data source, or a note saying it can't be answered yet.
Build the pipeline
Collection jobs, the SQL model, and a health check for each job.
Hand over
The dashboard, a short guide to every table, and the code in your own repository.
Other work
FMCG sales and distribution (Excel)
Sales workbook with three pivots on one shared cache, slicers, a per-rep report, and VBA macros that import a messy CSV (rejected rows logged with the reason) and export one PDF per sales rep.
Synthetic data, generated for the project.
Olist e-commerce (Power BI)
Four-page report (overview, customers, delivery, products) on the public Olist Brazilian e-commerce dataset, built as a PBIP project so every visual and measure is a text file under version control.
Tools
Have data that should be a dashboard?
Tell me where the data lives and what you want to know. I reply on LinkedIn.