Writing
Research & articles
Benchmarks, technical reports and long-form posts. Papers open right here in the browser.
Research · Sep 2026
SautiCivic Bridge: A Safety-Critical Speech Benchmark for Code-Switched Nigerian Civic Grievance Intake
Agoro & Akhuabe · Sahara CodeSwitch Africa Challenge 2026 · Technical Report
Benchmarks four ASR engines on code-switched Nigerian complaints and shows that WER hides "don"→"don't" polarity inversions (16.7–40% of complaints on global models, 0% on Sahara v2.5). The risk gate reaches 100% safe outcomes and 10/10 adversarial cases.
Research · May 2026
Personalised Recommendation via Semantic Retrieval and LLM Reranking
Agoro & Ajiboye · DSN x BCT Hackathon 3.0 · Task B
FAISS retrieval over 112,565 products blended with Bayesian popularity and Gemini reranking, across 5 Nigerian language modes. NDCG@10 0.4210, Hit@10 0.66.
Research · May 2026
User Modeling via Two-Agent Persona Simulation
Ajiboye & Agoro · DSN x BCT Hackathon 3.0 · Task A
Gemini Flash builds a persona dossier and Gemini Pro writes RAG-grounded reviews. RMSE 0.7071, BERTScore 0.8447; Nigerian conditioning lifts ROUGE-1 by 4.5%.
Article · Medium · Jul 25, 2026
I built QuantVault, a multi-agent runtime that deploys specialist AI personas from a single FastAPI backend
In progress, still being built (article only, no release yet). Lessons so far from building an agent ecosystem: LangGraph orchestration, YAML-swappable personas and tools, and a React dashboard for routing and monitoring.
Read on Medium
Article · Medium · Dec 22, 2025
The 2-Hour Gigabyte: Why Nigeria's Internet is a Trap
A data-driven look at telco pricing from 2021–2025: shrinking validity windows, diesel ghost-costs and 3G speeds at 5G prices, backed by a custom scraper and infrastructure reports.
Read on Medium