I build AI systems that turn raw data into intelligence. My work lives at the intersection of multi-agent orchestration, real statistical math, and production reliability — not demos, not wrappers.
I'm the founder of Busara, a data intelligence platform running 23 specialized TypeScript agents in a parallel DAG. Before Busara, I shipped IntelliFlow v2 — a 12-agent Python system that proved the architecture.
Based in Nairobi, building for Africa and beyond. Background in computer science, 5+ years shipping production systems, and a deep conviction that the next decade of AI will be multi-agent, distributed, and built by people who care about the math underneath.
From language primitives to payment rails — everything needed to ship an end-to-end AI platform.
Not a side project. Not a wrapper. A 23-agent system doing real math, in production.
A multi-agent data analysis platform with 23 specialized TypeScript agents running in a parallel DAG. Real math, not vibes:
Holt-Winters forecasting, Z-score / IQR / EWMA anomaly ensemble, OLS regression,
permutation importance, K-means clustering, and PII detection. LLM-powered narrative via GLM-4.6.
Flutterwave + Google Pay for monetization. PWA + native Android on the Play Store.
AI products, data science, and the work that taught me how to architect intelligence.
Created acla.io and its AI-assisted learning app, Tapi Learn. An adaptive learning platform that uses AI to personalize educational content, assess understanding, and generate practice materials in real-time.
Production time series forecasting toolkit — Holt-Winters, ARIMA, and ensemble models for financial metrics. 87.3% mean accuracy across 12 metrics. Built during work in the retirement benefits industry.
Multi-algorithm anomaly detection for financial transactions — Isolation Forest, autoencoder, and statistical ensemble. 94.2% detection rate with 2.1% false positives. Processes 50K+ transactions/second.
RFM analysis + K-Means + Gaussian Mixture Models for actionable customer segmentation. Identified 6 optimal segments with 89% month-over-month stability. 23% increase in targeted campaign response rate.
The direct predecessor to Busara. A 12-agent Python/Flask system for orchestrating data pipelines — proved the multi-agent DAG pattern before the TypeScript rewrite.
A timeline of building Busara and a career in data.
Founded Busara — 23 specialized TypeScript AI agents running in a parallel DAG, with GLM-4.6 LLM narrative, Flutterwave + Google Pay, voice input, SSE streaming, and native Android. Live in production.
Led data analytics and ML initiatives for retirement benefits — built forecasting models (87% accuracy), fraud detection systems (94% detection rate), and customer segmentation engines. Developed IntelliFlow v2, the 12-agent Python system that became Busara.
Led data infrastructure and analytics for Kenya's retirement industry regulator. Built data pipelines, automated compliance reporting, and deployed ML models for benefits fraud detection and member behavior analysis.
Started in benefits processing, grew into data analysis — built Excel/SQL dashboards for contribution tracking, automated claims processing reports, and identified data quality issues saving 200+ hours/month in manual reconciliation.
Open to collaborations on multi-agent systems, AI products, and ambitious African tech.
victor.ndunda@email.com