AIBackend

Corpus

A multi-agent pipeline that drafts academic manuscripts from real sources, and refuses to ship a citation it can't trace back.

Client
Universitas STEKOM
Year
2026

Internal system, not public

Corpus — main screen
7 StagesPipeline
6 StylesCitations

The problem

Getting a model to write text is easy. Getting citations that exist and actually support the sentence they're attached to is the hard part. General chatbots invent references without blinking.

What I did

Seven stages from retrieval to formatted output, ending in an integrity gate. Sources get stable short refs so citations are parsed straight from the model's markers. Researchers can ground drafts in their own Zotero library.

Stack

NestJS 11BunReact 19PostgreSQLDeepSeekZotero APIPandoc

More case studies

ENDContact

Want something like this built?

Hiring for a remote role, or have a system that needs untangling? My inbox is open.