AI Summarization Platform

Overview

The platform does not stop at a single model call; it turns summarization into a product flow completed by language/output checks, export, text-to-speech, and feedback.

Problem and goal

Producing a usable summary in the requested language requires more than sending a generation prompt. User history, feedback, and access boundaries also need management.

My role

• Developed Django REST summarization and user flows.
• Integrated TR/EN language/output checks, PDF export, and text-to-speech.
• Implemented JWT/refresh token, rate limiting, recent summaries, and feedback flows.

Solution

• Request and language checks ran before generation; output checks ran afterward.
• Summaries were retained in user history and made available through PDF and TTS formats.
• Helpful/not-helpful feedback was captured as a separate record flow.

Architecture

The system’s main components and data/decision flow are separated as follows.

Tech stack

Implementation decisions

• Decouple the model provider from the application contract.
• Apply rate limiting and identity flows around the AI call.
• Design language and output validation as separate stages.

Results and evaluation

• Built an application flow combining summarization, validation, export, TTS, history, and feedback.

Challenges and solutions

• Output consistency → add language and format checks to the generation pipeline.
• Resource use → bound calls through identity and rate limiting.

What I learned

The work reinforced that productizing an LLM feature also means designing validation, access, feedback, and output channels around the model response.