Frederico Luz Lisbon · —
← Work

Hexa Intelligence · Paid internship

Health-content portal

Turning source articles into health information people can read and question, with an editorial interface for reviewing the result.

Credit
Earlier engagement · Hexa
Role
Product engineering · paid internship
Status
Historical project · summer 2025
Stack
ReactSupabaseSource-aware generationRetrievalEditorial review
HealthPortal curator comparing source text with a simpler version beside an editorial assistant and approval controls
The original curator interface, run locally with synthetic sample text. Source and rewrite stay side by side; an editor can change, approve, or reject the result. No client data or live AI responses are shown.
More product screens (1)
HealthPortal source-entry dialog accepting an article URL or pasted text
Start with an article link or pasted text. This capture uses an example URL; no submission was made.

Make the source useful to the reader

Medical source material can be difficult to read, especially for readers working in a second language. During my paid internship at Hexa Intelligence, I built a health-content portal that accepted article links, produced more accessible health content, and offered a medical Q&A chatbot.

The product was not just a generator. A curator needed to compare the original with the rewritten version, make corrections, and decide whether the result was ready.

What I built

I implemented the source-to-content workflow and the React/Supabase editorial interface. The curator could inspect source and simplified text side by side, edit the rewritten content, and approve or reject it. The related retrieval work used heading-based chunks, embeddings, and source attribution to connect questions to relevant material.

Workflow diagram: a source article link is turned into a more accessible rewrite, which a curator compares side by side with the source, edits, and approves or rejects. In related retrieval work, source material is chunked by heading and embedded so a medical Q&A chatbot can answer with source attribution.
Generation hands off to a curator, who reviews the rewrite against its source before anything is approved.

The screen shown here is the original curator application, run locally with synthetic content. It demonstrates the comparison and review interface, not a current client deployment or a new live generation.

Decisions that shaped the work

  • Keep the original visible. Review is easier when the source and rewrite can be compared directly rather than accepted on the strength of fluent prose.
  • Let editors change the result. Manual edits and approval controls keep the workflow useful when a generated passage needs work.
  • Connect answers to source material. Heading-based retrieval and attribution make a medical Q&A interface more inspectable than an ungrounded chat response.

A distinct phase of the Hexa work

This was the summer 2025 internship, not the later BridgeSEO freelance contract. Related work in the same period explored citation mapping, image sourcing, and deterministic HTML assembly; those prototypes are not presented as a single fully integrated portal feature set.

The source-aware generation and editorial-review questions continued in BridgeSEO, where the problem expanded into a multitenant content-production platform with evaluation and cost controls.