Back to Work Archive
◈፬
CASE STUDY 04

Fedel AI

Contextual Video Subtitle Language Learning Engine

ROLESole Creator & Frontend/AI Engineer — Built the custom SRT file parser, vocabulary difficulty classifier, and interactive quiz generation pipeline.
TIMELINE2 Months (2024)
DOMAINAI & Systems
Fedel AI
፩THE CHALLENGE

The Problem Space

Traditional vocabulary flashcards lack organic emotional context. Language learners memorize words in isolation and struggle to grasp idiomatic cadence, colloquial phrasings, and situational humor present in authentic spoken dialogues.

፪SYSTEM ARCHITECTURE

Technologies & Tools

Next.jsTypeScriptTailwind CSSOpenAI APIWeb Audio API

Architectural Highlights

  • Lightweight client-side file parsing without server roundtrips
  • Accessible ARIA attributes on interactive dialogue tokens
  • Edge caching of common subtitle vocabulary models
፫ENGINEERING CHOICES

Key Architectural Decisions

DECISION 01

Deterministic Regex-Based SRT Tokenizer

Built a zero-dependency subtitle parsing tokenizer in pure TypeScript, normalizing timecode boundaries and extracting dialogue words at 60fps without performance drops.

DECISION 02

Semantic Word Frequency Pruning

Filtered out elementary words using frequency distribution lists, focusing generative AI prompts only on advanced idioms to conserve LLM token expenditure.

፬OBSTACLES & RESOLUTION

Challenges Faced

Encoding Variations in Multilingual SRT Files
HOW IT WAS SOLVED:

Engineered encoding detection fallbacks (UTF-8, UTF-16, ISO-8859) to prevent corrupted character rendering across international subtitle archives.

፭VERIFIED OUTCOMES

Results & Deliverables

01.Parsed full-length movie subtitle files in under 200ms on client devices.
02.Automated quiz generation workflow with zero manual question drafting required.
03.Accessible, high-contrast quiz interface built for rapid keyboard navigation.
NEXT CASE STUDYAi Mentor Plus
Read Next Case Study