Urgent.News

What's breaking now, across thousands of outlets.

AI

Tech expert unveils voice-first AI platform

Nigerian software engineer Kingsley Udoma launches Theresa AI, a voice-first platform designed to guide users through any topic in multiple languages. Disc Read More: https://punchng.com/tech-expert-unveils-voice-first-ai-platform/

Tech expert unveils voice-first AI platform

Nigerian software engineer Kingsley Udoma and his team have introduced a voice-first artificial intelligence system called Theresa AI. The system is designed to assist users in understanding any topic they may be struggling with, be it an assignment, a work problem, or a simple curiosity. Udoma and his co-founders, Jide Loye and Israel Akpan, unveiled the AI recently.

Udoma explained that the inspiration for Theresa AI came from observing people who often struggled to grasp information after reading or hearing it multiple times. He noted, "The problem was never a lack of content. After all, the internet is drowning in content. What I noticed was missing was somebody to actually talk to. So, instead of building another app to scroll through, my team built something that listens, responds, and thinks alongside you in real time."

What sets Theresa AI apart is its natural interaction with users. Users can ask questions out loud, and the AI responds with an answer out loud, eliminating the need for typing. Theresa AI can walk users through complex problems step by step, providing diagrams and flowcharts as it goes. The AI system also adapts to the user's tone, slowing down and exploring different angles if it detects confusion.

The co-founders of Theresa AI emphasized that the system understands and responds in several African languages, including Yoruba, Hausa, Igbo, French, and Spanish, making help accessible beyond just English speakers. Theresa AI is currently in testing phase with a wider rollout planned across Nigeria. Udoma aims to provide Nigerians with a 24/7 companion they can talk to, as one-on-one support is often unavailable in the country.

Udoma previously developed NounPaddi, an EdTech platform serving over 1,600 students at the National Open University of Nigeria, Lagos. There have been around half a dozen independent creators and more than 120 broader AI ventures that have launched conversational or localized large language models and AI agents inspired by ChatGPT in recent years. Victor, a Senior Correspondent at Punch Newspapers' Features Desk, reported this story, showcasing his experience in writing, research, and journalistic integrity.

Written by urgent.news from Punch Nigeria's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at punchng.com →

More in AI

We built a benchmark, then caught it strangling the models it was grading

A couple of day ago I posted about OmnisBench, our open benchmark for LLM routing, specifically our LLM Router OmnisRouter , and made a fuss about how you can re-grade every number yourself because we…

  • OmnisBench created to benchmark LLM routing with focus on OmnisRouter
  • Fresh split of LiveCodeBench excludes problems models could have trained on
  • Cheap model's output budget limited, causing poor performance on hard problems

Machine learning incorporating trabecular bone score improves lumbar fracture risk stratification in type 2 diabetes

Scientific Reports, Published online: 22 August 2026; doi:10.1038/s41598-026-67932-w Machine learning incorporating trabecular bone score improves lumbar fracture risk stratification in type 2…

  • Machine learning model using trabecular bone score improves lumbar fracture risk prediction
  • Study of 261 T2DM patients over 50 found significant improvement in fracture risk prediction
  • Logistic regression algorithm achieved highest performance with AUC of 0.818

Multi-label arrhythmia classification from Holter scatter plots using deep learning and clinically oriented thresholds

Scientific Reports, Published online: 22 August 2026; doi:10.1038/s41598-026-68106-4 Multi-label arrhythmia classification from Holter scatter plots using deep learning and clinically oriented…

  • Clinically oriented thresholds improved specificity to 0.885 but decreased macro-F1 score to 0.693.
  • Scatter plot-based deep learning could serve as efficient preliminary screening for Holter analysis.

A modular offline multilingual speech-to-speech translation system with confidence-aware ASR routing and language-aware TTS

Scientific Reports, Published online: 22 August 2026; doi:10.1038/s41598-026-67788-0 A modular offline multilingual speech-to-speech translation system with confidence-aware ASR routing and…

  • Modular offline system integrates language identification, routing, translation, and synthesis
  • Routing threshold of 0.70 lowers recognition errors in Indic languages
  • Mean model-only CPU latency of 25.41 seconds for offline multilingual speech translation

More from Friday 21 August →