
SERYNA - Emotionally Intelligent Voice AI Companion
A seamless fusion of LLaMA3, FAISS, and ElevenLabs delivering real-time empathetic interactions.

A document engine that turns any source material, including scanned pages, into summaries at multiple levels of detail and ready-to-use quizzes with answer keys, exportable as polished PDF or Word files.
Teachers at our client's organisation were spending significant time turning source material into study notes and test papers by hand. The client wanted this automated end to end: accept any document, including scanned pages, and produce summaries at different lengths along with ready-to-use quizzes and answer keys, exportable in common formats and able to work offline when needed.









Source material arrived as PDFs, Word files, and scanned pages that had to be handled consistently
The generation pipeline had to keep working without an internet connection
Generated quizzes needed varied question types, correct answer keys, and adjustable difficulty
Many scanned pages had no text layer and needed optical character recognition
Teachers needed summaries at different levels of detail, not a single fixed length
We built a document engine that extracts text from any supported file, including scanned pages via optical character recognition, into a single uniform format before any generation happens.
The engine splits content by chapter and section, then produces summaries at short, medium, detailed, or chapter-by-chapter levels from the same underlying material.
From that same content, the system generates multiple-choice, short-answer, and descriptive questions, each with an answer key and a chosen difficulty level, so teachers get a complete test paper, not just raw questions.
An AI router uses a cloud model when available and switches to a local model when offline, so generation never stalls. Finished summaries and quizzes export directly as polished PDF or Word files.
Teachers turn any document into study notes and a complete, answer-keyed assessment in minutes instead of hours.
PDFs, Word files, and scanned pages are all processed through the same consistent pipeline.
Summaries at four levels of detail and three question types cover a wide range of classroom needs.
Automatic online-to-offline fallback keeps the system working on the client's own hardware when required.

• By unifying document ingestion, multi-level summarisation, and automated assessment generation into a single pipeline with reliable offline fallback, this system turns hours of manual prep work into a task that takes minutes, without asking teachers to change how they source their material.
Explore other projects we've delivered

A seamless fusion of LLaMA3, FAISS, and ElevenLabs delivering real-time empathetic interactions.

Smart IoT solution for homes: experience real-time automation, voice control, and seamless device integration.

Transforming screen recordings into actionable performance insights with high-fidelity computer vision.