Jacob Rafal
Jacob Rafal
ProjectsAboutBlogComing SoonContact
Contact me
Jacob Rafal
GitHubLinkedInEmailCalendly

© 2025 Jacob Rafal

Jacob Rafal
Jacob Rafal
ProjectsAboutBlogComing SoonContact
Contact me
Back to Home

Tapat - AI Audio Agent for Philippine Elections

An AI-powered audio conversational agent providing real-time information and fact-checking for the 2024 Philippine elections

WhisperGPT-4FastAPILangChainReactPythonSpeech Recognition

Demo Video

Overview

Tapat is an innovative AI audio agent designed to democratize access to election information in the Philippines. Built during the 2024 election season, it provides voters with instant, accurate information about candidates, platforms, and electoral processes through natural voice conversations.

The system leverages state-of-the-art speech recognition and large language models to understand queries in both Filipino and English, making civic information accessible to all voters regardless of their technical literacy.

Key Features

  • ✓Real-time audio conversation with sub-2 second response time
  • ✓Bilingual support (Filipino and English) for inclusive access
  • ✓Fact-checking capability to combat misinformation
  • ✓Coverage of 100+ candidates and their platforms
  • ✓RAG system for accurate, sourced information retrieval

Technical Architecture

Speech Processing

  • • Whisper for transcription
  • • Multi-language detection
  • • Noise reduction

AI Processing

  • • LangChain orchestration
  • • GPT-4 for responses
  • • RAG for accuracy

Backend

  • • FastAPI server
  • • WebSocket connections
  • • Redis caching

Social Impact

🗳️

Voter Education

Empowering citizens with accurate election information through accessible AI

🛡️

Combat Misinformation

Real-time fact-checking to ensure voters receive verified information

🌏

Inclusive Access

Breaking language and technology barriers for all Filipino voters

Technical Challenges Solved

Low-latency Audio Processing

Achieved sub-2 second response times through optimized pipeline design, WebSocket streaming, and intelligent caching strategies.

Multilingual Understanding

Implemented code-switching detection to handle mixed Filipino-English conversations naturally, crucial for authentic Filipino communication patterns.

Information Accuracy

Built a comprehensive RAG system with verified sources to ensure all election information provided is accurate and up-to-date.

Key Achievements

<2s
Response Time
100+
Candidates Covered
Bilingual
Fil/Eng Support
24/7
Availability
LinkedIn PostGitHub Profile

Interested in building AI solutions for social good?

Let's Collaborate
Jacob Rafal
GitHubLinkedInEmailCalendly

© 2025 Jacob Rafal