Available for select engagements · Nairobi 🇰🇪 & remote

Victor NdundaProduction AI systems engineerNairobi, KE

AI systems that survive production.

I design, build, and run RAG knowledge tools, agents, forecasting, and fraud-detection systems — with the evaluation and observability to prove they work. Pilot to production, in English, Français, and Kiswahili.

Book a 30-Minute AI Systems Assessment See the systems I've shipped
DWG № VND-28 · PRODUCTION LINE REV 12 — 2026
LISTINGS 87+ LIVE DOCS TITLES TXNS 1K LABELLED EVAL LOOP — 281 DEFECTS CAUGHT IN PUBLIC REVIEW INGEST 87+ LISTINGS · 12 COUNTIES EMBED EN · FR · SW RETRIEVE HYBRID + RERANK GENERATE CITED · ~33µs/SCORE EVAL 84% RECALL · 1.37% FPR USERS 11K+ LEARNERS · 38 COUNTRIES
13 SYSTEMS DESIGNED, BUILT & SHIPPED · TAP A STAGE FOR ITS EVIDENCE
13AI systems designed, built & shipped
11K+learners on a shipped client product
EN·FR·SWmultilingual, low-bandwidth delivery
↓
Work With Me

Three ways to work together.

Every engagement begins with a diagnosis — never with a stack recommendation. Full service detail with USD & KES pricing lives on the services page.

Diagnose

Needs Assessment

Know exactly what to build — and what to leave alone.

Best for: teams with little or no AI in production — or shipped AI that breaks, drifts, or burns money.

Start with an assessment
  • Workflow map & data-readiness review
  • ROI-ranked opportunity map (5–10 use cases)
  • Build / no-build recommendation
  • Reliability mode for shipped systems
$2,500 · KES 200K · 1–2 weeks
Most commissioned

Starter Build

Ship your first real AI tool, not a prototype.

Best for: SMBs, solo founders, NGOs — one workflow with an owner.

Discuss a starter build
  • WhatsApp or web assistant + 1 automation
  • Knowledge base with citations (50 docs)
  • English + Swahili out of the box
  • Docs, handoff & 30-day support
$12,000 · KES 960K · 2–3 weeks
or 3 instalments of $4,200 · care $1,400/mo
Commission

Big Projects

Production AI programs that run the business.

Best for: companies commissioning serious systems — mid-market to enterprise.

Scope a big project
  • RAG platforms & multi-agent orchestration
  • CRM/ERP integrations + predictive analytics
  • Compliance-reviewed, SLA-backed (99.5%)
  • Custom-scoped after a Needs Assessment
From $35,000 · KES 2.8M · 6–24 weeks
Explore services & pricing
Case Studies

Case studies, with numbers.

Four flagship engagements — each with the constraint that made it hard and the evidence that it works. Two have full written case studies; the full archive — 13 projects spanning client builds, platforms, data-science work, and open-source tools — is on the case-studies page. Explore all projects →

Client engagement · Real estate

Keja AI · CHACADOM INVESTMENTS

A 7-stage real-estate intelligence platform — live in public trial with 87+ listings across 12 Kenyan counties.

Verified titles & documents, investment scoring on every listing, market data, tokenized assets, and the "Ask Keja AI" assistant — across dedicated workspaces for 8 stakeholder types, in English, Kiswahili, and Français. Nominated for an Africa AI Future Award.

87+Listings (public trial)
12Kenyan counties
70Trust Score ≥ 80
Read the full case study → Live Platform ↗
Client engagement · EdTech

Tapi Learn · ACLA

A multilingual AI learning product that reached 11,000+ learners across 38 countries.

Africa-built climate-leadership platform. The AI Career Coach, the multilingual NLP pipeline (English · Français · Kiswahili), and personalized learning recommendations had to coach real learners in production.

11K+Learners (client-reported)
38Countries
4.8★Avg rating (client-reported)
Read the full case study → Live Product ↗
Workflow automation · SaaS

Verxlite

A sales & ops AI agent that automates follow-ups, CRM logging, and approvals — with its reliability proven in a public 281-defect code review.

Multi-tenant SaaS with real OAuth (Google, HubSpot), encrypted tokens, Celery workers, and Langfuse observability. Every P0/P1 finding was remediated within one week of the review.

281Defects publicly reviewed
83Backend tests passing
GitHub ↗ Code Review Automate a workflow →
Fintech · Regulated finance

Fraud Detection System

Real-time transaction scoring at 84% recall and a 1.37% false-positive rate — ~33 µs per decision.

Seven statistical detectors fused by a weighted-vote ensemble, calibrated by logistic regression, overlaid with a deterministic rules engine — every score explainable and auditable. Validated on 1,000 labelled transactions; methodology published with the code. Designed with controls relevant to PCI-DSS, PSD2, and GDPR automated-decision requirements — formal compliance is deployment-specific.

84%Recall
1.37%False positives
~33µsPer score
GitHub ↗ Discuss a fintech system →
How It Works

Diagnose. Prove. Ship.

Pilot-to-production is where most AI dies. This engagement model is designed around getting past it.

Diagnose

Define the user, the workflow, the constraint, and the success metric. No stack choices before this — most "AI projects" fail here, not in the code.

Prove

Build the smallest useful slice and test it against real scenarios with a real evaluation set. Numbers or it didn't happen.

Ship

Harden, document, observe, and hand over a system your team can run — with the dashboards to prove it stays healthy.

Fixed scope, fixed price.What we signed is what you pay. No surprise invoices, ever.
You own everything.Code, prompts, data, infrastructure accounts — transferred at handover.
Milestone delivery.Big builds run 30/40/30 — you see working software at every gate.
No lock-in.Leave anytime — with a running, documented system. Care is a service, not a hostage.
About

Hi, I'm Victor.

I'm an AI engineer based in Nairobi, Kenya. I started in data science for fintech and financial services — forecasting, fraud detection, customer segmentation — and moved to production AI full-time when generative models matured: multi-agent platforms, RAG pipelines, on-device ML, and multilingual assistants.

What I actually do is take technically difficult AI ideas and make them work under real constraints — $80 phones, Swahili and French delivery, regulated finance, weak connectivity. The systems below — and the client products above — are the evidence. I also write the engineering down: 10 long-form articles, 22,000+ words, because work you can't explain is work you can't trust.

I take on select client work — audits, sprints, and full builds. If you have a problem that needs real engineering, let's talk.

Currently building: Keja AI public trial (award-nominated) · FirstLight Geo-AI demo · AgentReplay 1.x — write-ups in Insights.

View full resume Download PDF

End-to-End Builds

From scoping audit to production deployment. Full-stack: Next.js, FastAPI, Flutter, PostgreSQL, Docker, Kubernetes.

Multi-Agent & RAG

Multi-agent DAG orchestration, RAG with pgvector/Pinecone, hybrid search, reranking. LangChain, LangGraph, CrewAI.

Statistical Foundations

Holt-Winters forecasting, GARCH/VaR, K-Means++, OLS regression, Granger causality — implemented from scratch.

Built for Africa

Offline-first, low-bandwidth, Swahili-capable. Systems that work on $80 phones and over SMS/USSD.

The Work

Shipped systems, latest first.

Live client platforms lead, open-source foundations follow — every system designed, built, evaluated, and maintained end-to-end. The full 13-project archive is on the case-studies page.

Client Platform · Award Nominee

Keja AI

Real-estate intelligence and trust infrastructure for African property — one ecosystem covering the full transaction lifecycle. Verified titles & documents, investment scoring on every listing, market data, tokenized assets, and an "Ask Keja AI" assistant, with dedicated workspaces for buyers, sellers, landlords, tenants, investors, developers, banks, and institutions. Built for CHACADOM INVESTMENTS; live in public trial at keja.app. Nominated for an Africa AI Future Award — ceremony 27 Nov 2026, Nairobi.

Next.jsAI AssistantInvestment ScoringTitle VerificationMarket DataTokenizationTrust Score
Live Platform ↗ Read the case study → Discuss a similar build →
7-Stage Transaction Lifecycle
Discover1
→
Verify2
→
Analyse3
→
Finance4
→
Invest5
→
Transact6
→
Manage7
→
Trust Score≥80
87+Listings live (Oct 2026)
12Kenyan counties
EN·SW·FRLanguages
Client Platform · EdTech

Tapi Learn

The multilingual AI learning product behind ACLA's climate-leadership academy — a 24/7 AI Career Coach (CV reviews, interview prep, career advice), a personalized recommendation engine, and an EN · FR · SW NLP pipeline coaching real learners in production: 11,000+ learners across 38 countries (client-reported, Oct 2026) at a 4.8/5 average rating.

11K+Learners (client-reported)
38Countries
4.8★Avg rating (client-reported)
Multilingual NLPAI Career CoachRecommendationsEN·FR·SWMobile
Live Product ↗ Read the case study → Discuss a similar build →
Geo-AI · Disaster Response · New

FirstLight

A Geo-AI decision layer for the first six hours after a disaster — damage & flood detection fused with road networks, population grids, and critical facilities: Uber H3 need-score zoning, Dijkstra routing on the post-event road graph, hospital isochrones, isolation analysis, a human-in-the-loop review queue, and hash-verified situation reports. Fully client-side public demo — every artifact simulated and provenance-labeled.

Next.jsMapLibre GLUber H3Geo-AISARTypeScript
Live Demo ↗ GitHub
Open-Source · Agent Reliability

AgentReplay

Deterministic replay and counterfactual debugging for AI agents. Captures every LLM call, tool response, clock read, and RNG draw — then replays the run bit-exact with zero model calls. 0.67% overhead, 100% reproduction fidelity, 99 tests. LangGraph-first. Backed by a research whitepaper.

PythonLangGraphOpenAI SDKAnthropic SDKDeterministic ReplayCounterfactualCI Regression
GitHub ↗ Whitepaper
Enterprise · Regulated Finance

Fraud Detection System

A real-time transaction fraud scoring engine in Go — 7 statistical detectors (Z-Score, IQR, Velocity, Geo-Distance, Device Fingerprint, Merchant Risk, Behavioral Anomaly) fused by a weighted-vote ensemble, calibrated by logistic regression, and overlaid with a deterministic rules engine. Scores in ~33 µs per transaction. Full case-management workflow, RBAC auth, Prometheus metrics, OTel tracing, and Slack alerts. 84% recall, 1.37% false-positive rate on 1,000 labelled transactions. Designed with controls relevant to PCI-DSS, PSD2, and GDPR Article 22 — formal compliance is deployment-specific.

~33µsScore latency
84%Recall
1.37%False positives
7Detectors
Go 1.22Gin7 DetectorsWeighted EnsembleLogistic CalibrationRules EngineCase ManagementRBAC + JWTPrometheusOpenTelemetrySlack AlertsKubernetes
GitHub ↗ Architecture
Sales & Ops Automation

Verxlite

A universal AI workflow agent for sales & ops teams. Automates post-meeting followups, lead assignment, support triage, approval routing, and weekly summaries across Gmail, Google Calendar, and HubSpot. Multi-tenant SaaS with real auth, OAuth, encrypted tokens, Celery workers, and Langfuse observability. Shipped with a transparent 281-defect code review and full P0/P1 remediation in one week — 83 backend tests passing.

PythonFastAPICeleryNext.js 14SQLAlchemy 2AlembicGoogle OAuthHubSpot OAuthLangfuseFernet
GitHub ↗ Code Review
Data Intelligence Platform · Live

Busara AI

A multi-agent data intelligence PWA running a 6-stage parallel DAG pipeline (v7.0) — Ingest → Engineer → Detect → Forecast → Infer → Report. Statistical algorithms implemented from scratch: Holt-Winters, OLS, K-Means++, Granger causality, GARCH. Live at busaraai.com with a Free → Pro → Team → Enterprise tier (99.9% SLA target); the IntelliFlow foundation is open-source.

TypeScriptNext.js 15Multi-Agent System6-Stage DAGGLM-4.6PostgreSQLPWAFlutterwave
Live Platform ↗ GitHub
Agricultural Intelligence

KilimoPRO

AI-powered agricultural intelligence for smallholder farmers across 8 IGAD countries — a 300M+ person market. On-device crop disease detection (TFLite) for 15 crops, market forecasting with GARCH/VaR, multi-LLM advisory council in Swahili + English, 9 free keyless API endpoints, SMS webhook for feature phones. $0/month to operate on free data sources.

Next.js 14Drizzle ORMTFLiteOpen-MeteoFAOSTATICPACMulti-LLM CouncilSMSOffline-First
GitHub ↗ Case Study

IntelliFlow (Busara AI)

33-agent self-evolving multi-agent data intelligence platform. 7-stage pipeline with trajectory-based learning inspired by the AReaL paper. Turborepo monorepo, Next.js 15.

TypeScriptMulti-AgentSelf-EvolvingTurborepo
GitHub →

Financial Forecasting Suite

Production-grade time series forecasting toolkit. Holt-Winters, ARIMA/SARIMA, Prophet-style decomposition, ensemble forecasting with auto model selection. 87.3% mean accuracy across 12 financial metrics.

PythonHolt-WintersARIMAEnsemble
GitHub →

Customer Segmentation Engine

RFM analysis + K-Means + Gaussian Mixture Models for actionable customer segmentation. 6 optimal clusters, 89% month-over-month stability, 23% lift in campaign response rate. Built for financial services.

PythonK-MeansGMMRFM
GitHub →

Eco Africa Awards

Celebrating climate leadership across Africa — an initiative of ACLA. Built with Next.js, Three.js, and AI. Live at ecoafricaawards.com.

TypeScriptNext.jsThree.jsAI
GitHub →

Hazina Assist

Early-stage project now in development — details will be published at the first public milestone.

In Development
Follow the archive →
Explore All Case Studies
Python•TypeScript•JavaScript•Dart•SQL•Bash•GPT•Claude•Gemini•GLM-4.6•Kimi K2•MiMo•Mistral•Qwen•DeepSeek•Llama•LangChain•LangGraph•LlamaIndex•CrewAI•AutoGen•Multi-Agent DAG•pgvector•Pinecone•Weaviate•Hybrid Search•Reranking•TensorFlow•TensorFlow Lite•scikit-learn•Holt-Winters•GARCH•VaR•K-Means++•OLS•Granger Causality•Next.js•React•Node.js•FastAPI•Flutter•PostgreSQL•Redis•Docker•Kubernetes•GCP•Earth Engine•CI/CD•Git/GitHub•Evals•Observability•
Tech Stack

Tools I build with.

Click a category to filter. Every tool here is something I've shipped to production — from multi-agent DAGs to on-device ML.

Python
Language
TypeScript
Language
JavaScript
Language
Dart
Language
SQL
Language
Bash
Language
Go
Language
GPT (OpenAI)
Foundation Model
Claude (Anthropic)
Foundation Model
Gemini (Google)
Foundation Model
GLM-4.6/4.5 (Zhipu)
Foundation Model
Kimi K2 (Moonshot)
Foundation Model
MiMo (Xiaomi)
Foundation Model
Mistral
Foundation Model
Qwen (Alibaba)
Foundation Model
DeepSeek
Foundation Model
Llama (Meta)
Foundation Model
LangChain
Agent Framework
LangGraph
Agent Framework
LlamaIndex
Agent Framework
CrewAI
Agent Framework
AutoGen
Agent Framework
Custom Multi-Agent DAG
Orchestration
pgvector
Vector DB
Pinecone
Vector DB
Weaviate
Vector DB
Hybrid Search (BM25 + Vector)
Retrieval
Reranking
Retrieval
Chunking Strategies
Retrieval
TensorFlow
Deep Learning
TensorFlow Lite
On-Device ML
scikit-learn
Machine Learning
Holt-Winters
Forecasting
GARCH / VaR
Quant Finance
K-Means++
Clustering
OLS Regression
Statistics
Granger Causality
Statistics
Next.js
Framework
React
Framework
Node.js
Backend
FastAPI
Backend
Flutter
Mobile
PostgreSQL
Database
Redis
Cache
Docker
Containers
Kubernetes
Orchestration
Google Cloud Platform
Cloud
CI/CD
DevOps
Git / GitHub
Version Control
Google Earth Engine
Geospatial
Evals & Observability
LLM-as-Judge, Tracing
Open Source

Recent GitHub activity.

Live from my GitHub — what I've been pushing lately. Updated in real-time via the GitHub API.

Loading recent repositories…
View all repos on GitHub
Insights

Thinking it through, in public.

10 long-form articles on production AI — agent reliability, RAG, offline-first systems, statistical engineering. Three that buyers tend to start with:

Reliability

Why agent demos fail in production — and how deterministic replay fixes it

AgentReplay reproduces agent failures bit-exact with zero model calls — the research basis behind the Reliability Audit offer.

Read the article → Forecasting

The statistical engines behind honest forecasting

Holt-Winters, GARCH, OLS — implemented from scratch. Why real math matters when forecasts drive decisions.

Read the article → Offline-first

Offline-first AI for 8 IGAD countries

On-device disease detection, SMS fallbacks, and Swahili delivery — AI that works where connectivity doesn't.

Read the article →
All 10 articles →
Contact

Let's build something.

Direct channels for existing conversations — new ones usually start with the assessment call below.

Email
mututandunda@gmail.com
WhatsApp · Chat or Call
+254 724 346 971
GitHub
@gadda00
LinkedIn
Victor Ndunda
Save Contact
victor-ndunda.vcf · one tap
Free · 30 minutes · No pitch ·

Book a 30-Minute AI Systems Assessment

Bring one workflow, bottleneck, or AI idea. You'll leave with an honest read on what's worth building, what's risky, and the smallest useful first step — even if that step is "not yet."

Book the Assessment See how we'd work together

Prefer async? Email me directly — or find me on WhatsApp. Not ready to talk? Score your AI readiness in 2 minutes — free, private, no email needed.

Book an Assessment