Backend · data · machine learning · AI inside
Nexova Platform
An operations platform for a (fictional) HR and recruiting firm: public site, back office, background jobs, data pipelines and a machine-learning revenue forecast.
- Role
- Backend, data and ML engineering
- Status
- AI Engineering capstone · runs in Docker
- Link
- Demo on request
- 2.61%
- average monthly forecast error
- 23 / 24
- months inside the predicted range
- 80
- commits · 23 merged pull requests
- 10
- automated test files
The problem
A growing company needs more than a website: jobs that run reliably overnight, slow tasks that don’t freeze the app, clean data, and an honest answer to “what will revenue look like next year?”
What I built
- ✦Revenue forecast model trained on 8 years of sales and tested on the next 2: 2.6% average monthly error, with 23 of 24 months inside the predicted range
- ✦Background task queue with automatic retries, timeouts and a dead-letter queue, so nothing fails silently
- ✦A nightly export job that is safe to re-run and backs data up before processing it
- ✦Event tracking across the front end with batching and retry, feeding a reporting dashboard
- ✦REST API with secure login, inventory and incident tracking, and a recruiting pipeline tracker
- ✦Bilingual (English / Spanish) public site, automated tests, and a one-command Docker setup
Where the AI lives
A Random Forest model forecasts monthly revenue, built with guardrails against the most common forecasting mistake — accidentally letting the model peek at the future — and tests that enforce it.
What it shows I can build for you
If your business runs on spreadsheets and guesswork, I can build the pipes underneath: scheduled jobs, dashboards, clean data and forecasts you can actually plan around.
Built with
- Python
- FastAPI
- PostgreSQL
- Celery
- Redis
- Prefect
- scikit-learn
- pandas
- Next.js
- Docker
- pytest