Oscar Ndugbu — Principal Full-Stack Engineer · AI Infrastructure

AVAILABLE · OPEN TO WORK· Updated May 2026

Full-Stack · Java · React Native · Next.js 15 · AI Systems · Fintech

Production systems that stay alive when it matters most — compliant, fast, and relentlessly reliable. Built under Lagos constraints. Deployed to global standards.

System IDOscar Ndugbu

Staff+

Architecture StackNode.js · Java · ML Infra · Next.js 15

Lagos · UTC+1

Shipped in LagosRunning globallyBattle-tested in audit season

TaxBridge · NRSAudit live

4h → 15min

Tax filing timeCompressed a live tax workflow under real audit-season pressure.

Fault Tolerance · ArchitectureVerified

Zero-Drop

System ResiliencySelf-healing queues and idempotent processing. Designed for zero data-loss.

API surface · p99Stable

sub-150ms

API responseFast under real user load, with Lagos network reality in the loop.

SabiScore · opsImproved

45% MTTD

Incident detectionDetection tightened before incidents could turn into user pain.

Tell me your constraints

See what shipped

I respond within 24 hours — usually faster.

Download CV

Or read how the 2am constraint became the design system

System IDOscar Ndugbu

Staff+

Architecture StackNode.js · Java · ML Infra · Next.js 15

Lagos · UTC+1

TaxBridge: 4h → 15min filing · NRS compliance · zero data-loss record.SabiScore: ensemble ML inference · 45% MTTD improvement · zero-drop queue design.SwarmXQ: self-improving agent fleet · checkpoint recovery · zero manual tuning.Shipped in Lagos · Running globally · Battle-tested in audit season

LIVE IN PRODUCTION

Three ML models working in ensemble catch system problems 45% faster than traditional alerting — before users notice anything is wrong. Production reality, not staging theater.

3 live production systemsFederal-scale delivery · 40M students99.9%+ sustained uptimeNRS · NDPC compliantLagos → Global

AVAILABLE · OPEN TO WORK· Updated May 2026

Full-Stack · Java · React Native · Next.js 15 · AI Systems · Fintech

Production systems that stay alive when it matters most — compliant, fast, and relentlessly reliable. Built under Lagos constraints. Deployed to global standards.

System IDOscar Ndugbu

Staff+

Architecture StackNode.js · Java · ML Infra · Next.js 15

Lagos · UTC+1

Shipped in LagosRunning globallyBattle-tested in audit season

TaxBridge · NRSAudit live

4h → 15min

Tax filing timeCompressed a live tax workflow under real audit-season pressure.

Fault Tolerance · ArchitectureVerified

Zero-Drop

System ResiliencySelf-healing queues and idempotent processing. Designed for zero data-loss.

API surface · p99Stable

sub-150ms

API responseFast under real user load, with Lagos network reality in the loop.

SabiScore · opsImproved

45% MTTD

Incident detectionDetection tightened before incidents could turn into user pain.

Tell me your constraints

See what shipped

I respond within 24 hours — usually faster.

Download CV

Or read how the 2am constraint became the design system

System IDOscar Ndugbu

Staff+

Architecture StackNode.js · Java · ML Infra · Next.js 15

Lagos · UTC+1

LIVE IN PRODUCTION

Three ML models working in ensemble catch system problems 45% faster than traditional alerting — before users notice anything is wrong. Production reality, not staging theater.

Projects

Built to survive
real constraints.

4-hour tax filings compressed to 15 minutes. 99.9%+ uptime under ensemble ML inference. AI agents that improve themselves between runs. All of it shipped from Lagos. All of it running in production.

Compliance Platform · FintechCASE STUDY

TaxBridge

4 hours of Nigerian SME tax filing compressed to 15 minutes — NRS-integrated, audit-ready, zero data-loss.

4hrs → 15min filing
sub-150ms under load
95% test coverage
zero data-loss record

WHYNRS audit scrutiny demands proof that tenant data cannot cross-contaminate — RLS enforces this at the database engine level, not the application layer

Constraint: NRS API rate limits: 30 req/min per TIN. BullMQ queue must manage burst filing windows without client-visible failure.

Fastify 5Java 17Spring Boot 3PostgreSQL 15 RLSRedis 7+6 more

Full brief ↓

Full tax compliance workflow automation for Nigerian small businesses — VAT, withholding tax, and annual returns. PostgreSQL RLS isolates each tenant at the database level. Real-time calculations under <150ms at load. Idempotent BullMQ job queue ensures no submission is ever double-processed — even through mid-request server failure. Hash-chained immutable audit trail. 95% test coverage.

BECAUSE

NRS audit scrutiny demands proof that tenant data cannot cross-contaminate — RLS enforces this at the database engine level, not the application layer

CHOSENPostgreSQL Row-Level Security for multi-tenancy

OVERApplication-layer tenant filtering

Read case study

ML Platform · ObservabilityLIVE

SabiScore

99.9%+ uptime. 45% MTTD improvement. Ensemble ML that alerts before users notice.

30% inference latency reduction
99.9%+ uptime
45% MTTD improvement
Prometheus 90-day proof

WHYSub-50ms event fan-out at sustained load with dead-letter recovery — impossible with polling under concurrent sessions

FastAPIXGBoostLightGBMCatBoost+4 more

Read the case study Live demo View source

AI Agent Platform · OrchestrationLIVE

SwarmXQ

Self-improving multi-agent operator platform — autonomous evolution layer, live fleet dashboard, production-grade reliability under resource constraints.

self-improving agent fleet
live orchestration dashboard
checkpoint-based fault recovery
zero manual tuning cycles

WHYManual tuning cannot adapt to novel inputs at scale — autonomous evolution scores strategies against real outcomes and rewrites low performers between runs, compounding quality without engineering intervention

PythonFastAPIOllamaNext.js 15+4 more

Read the case study View source

Blockchain · ZK PrivacyCASE STUDY

Hashablanca

ZK proofs for document integrity verification — confidentiality and verifiability as simultaneous properties, not a tradeoff.

ZK proof integrity layer
4 chain networks
4GB+ file processing

BECAUSEDatabase timestamps are mutable — ZK proofs provide cryptographic verifiability of document existence and integrity without exposing content

CHOSENGroth16 ZK proofs with off-chain proving

Circom 2snarkjs (Groth16)SolidityFastAPI+3

Read case study View source

Federal Infrastructure · Data EngineeringCASE STUDY

UBEC Data Pipeline

Abuja HQ · Federal-scale education data for 40 million Nigerian students across 36 states — probabilistic deduplication, per-state retry semantics, ministry-grade reporting.

36 state sources ingested
<2% dedup false-positive rate
per-state retry semantics

BECAUSEState submissions use inconsistent school name spellings across ministries — exact match misses 15–20% of true duplicates in validation runs

CHOSENBlocking + probabilistic record linkage (dedupe.io)

Python 3.11Apache AirflowpandasPostgreSQL+3

Read case study

How these systems perform in production →

Production record

Shipped systems.
Verified outcomes.

Three live platforms. One federal engagement. Every metric traceable to a deployed codebase.

COMPLIANCE PLATFORM · FINTECH

4h → 15minFiling time

Tax filing compressed. Zero data-loss record.

BullMQ absorbs NRS API rate limits (30 req/min per TIN) without client-visible failure. PostgreSQL RLS isolates each tenant at the database engine level — not the application layer. Immutable hash-chained audit trail. 95% test coverage.

Chosen: RLS at engine level, not application-layer filtering — NRS audit scrutiny demands proof that tenant data cannot cross-contaminate.

ML PLATFORM · OBSERVABILITY

99.9%+90-day uptime

Ensemble ML. Engineers alerted before users notice.

XGBoost, LightGBM, and CatBoost inference with real-time model quality monitoring. ~30% inference latency reduction via Redis caching. 45% MTTD improvement over reactive alerting baseline. Prometheus 90-day proof window.

Chosen: FastAPI + Redis Pub/Sub — sub-50ms event fan-out at sustained load, impossible with synchronous polling under concurrent sessions.

AI AGENT PLATFORM · ORCHESTRATION

ZeroManual tuning cycles

Self-improving fleet. Checkpoint recovery.

Triadic LLM routing under 8GB RAM — Phi-4-mini for task routing, DeepSeek-R1 for multi-step reasoning, Qwen2.5-Coder for generation. Agents evolve their own task strategies between runs. Failed sub-tasks restart from last consistent checkpoint, not from scratch.

Chosen: Autonomous evolution layer over static agent configs — manual tuning cannot adapt to novel inputs at scale.

FEDERAL INFRASTRUCTURE · DATA

36State sources

40M Nigerian students. <2% deduplication error.

Batch ingestion for the Universal Basic Education Commission across all 36 Nigerian states. Probabilistic record linkage (dedupe.io + PostgreSQL) where exact-match alone misses 15–20% of true duplicates. Per-state DAG tasks — one late submission does not block reporting for the other 35.

Chosen: Probabilistic deduplication over exact-match — state submissions use inconsistent school name spellings across ministries.

Tools extracted from these systems and open-sourced →

Open Source

Infrastructure for problems nobody packaged yet.

Four production-hardened packages from the fintech trenches — each solving a gap that general-purpose libraries don’t address. Install in minutes.

Four packages extracted from production TaxBridge and SabiScore — solved for real, then published.

Node.js · PostgreSQL

Production-grade

pg-tenant

Row-Level Security at the engine — not the app. Even when application bugs exist, one tenant's records are mathematically invisible to another. Production-deployed in TaxBridge, where a single data-leak event means regulatory audit.

Signal: DB-layer tenant isolationEvidence: Shipped in TaxBridge

View on GitHub

Fintech · Compliance

NRS · GDPR

audit-chain

Retroactive tampering becomes mathematically detectable. Every log entry is cryptographically bound to the previous — any edit breaks the chain instantly. Built for NRS and GDPR trails where proof of integrity is non-negotiable.

Signal: Tamper-evident log chainEvidence: NRS · GDPR audit trails

View on GitHub

AI · DevOps

AI-powered

node-debug-llm

Incident triage in minutes, not hours. Streams live system logs and traces to an AI model, returning a ranked plain-English list of likely root causes — with full context, not just stack traces.

Signal: Ranked root-cause suggestionsEvidence: Live traces + logs input

View on GitHub

AI · Agent Orchestration

Zero cloud

llm-dispatch

Triadic model routing extracted from SwarmXQ. Routes tasks to the right local GGUF model — Phi-4-mini for classification, DeepSeek-R1 for reasoning, Qwen2.5-Coder for code — with fallback chains and zero cloud egress.

Signal: Triadic local model routingEvidence: Zero cloud egress

View on GitHub

4production packages· npm + pip

15+merged contributions· XGBoost · scikit-learn

MITopen license· all packages

All packages are publicly auditable → GitHub

The stack that makes these possible →

Skills

The stack behind systems that hold.

62 tools deployed under NRS audit pressure (Nigeria's federal tax compliance mandate), 99.9%+ uptime targets, and multi-tenant isolation constraints. Every skill maps to a decision made in production — not a tutorial, not a certification exercise, not a side project.

4h → 15min

tax filing time

TaxBridge · Fastify 5 · PostgreSQL 15 RLS

99.9%+

sustained uptime

SabiScore · XGBoost · LightGBM · Prometheus

Self-improving

AI agent fleet

SwarmXQ · Ollama GGUF · Checkpoint Recovery

Core production stack

Default view is intentionally curated for fast trust. Open the full explorer only when you want deep implementation coverage across all 62 tools.

ExpertTypeScript stricttaxbridge · portfolio

ExpertReact 19portfolio · sabiscore

ExpertNext.js 15portfolio · sabiscore

ProReact Native / Expo SDK 54taxbridge

ExpertFastAPIsabiscore · taxbridge

ExpertJava 17+taxbridge · ubec

+6 more in the full explorer below.

Production system × skill lineage

TaxBridge

TypeScript strictJava 17+Spring Boot 3PostgreSQL 15 RLSFastify 5React Native / Expo SDK 54BullMQ (idempotent queues)Row-Level Security (RLS)

SabiScore

XGBoostLightGBMCatBoostFastAPIRedis 7 / Pub-SubPrometheus + GrafanaPython 3.11+Feature Engineering

SwarmXQ

Next.js 15FastAPITriadic LLM RoutingMulti-Agent OrchestrationCheckpoint-based RecoveryPython 3.11+Ollama (GGUF dispatch)pgvector

Full explorer · 62 tools · 8 pillars

Expanded layer for CTO and engineer-level scrutiny. Optional by design to preserve reading momentum for recruiters and founders.

These 62 skills map to three live systems →

About

Federal scale.
Production ML.
Lagos → Global.

Built the digital learning infrastructure for 40 million Nigerian students at UBEC (Abuja HQ). Then shipped three production systems from Lagos that hold at 2am.

4h → 15minTax filing time

99.9%+Sustained uptime

3Production platforms

The UBEC work meant no staging mirror, no padded support layer, and no room for vague ownership at federal scale. That operating constraint carried into every system after it. First-principles thinking, explicit failure modes, architecture decisions written so the next engineer can read the reasoning long after deploy.

TaxBridge: React Native / Expo 54 mobile app, Fastify 5 API, PostgreSQL 15 RLS for tenant isolation. SabiScore: ensemble ML inference (XGBoost, LightGBM, CatBoost) behind a Next.js 15 dashboard — 99.9%+ uptime across a 90-day Prometheus window. SwarmXQ: self-improving AI agent fleet with live ops visibility.

Fifteen-plus upstream contributions merged, four cloud certifications, and public code that outlives any single job title. Constraint is the credential. The work is the record.

MobileReact Native

WebNext.js 15

UIReact 19

LanguageTypeScript

APIFastAPI

DataPostgreSQL 15

MLPython 3.11+

CacheRedis 7

Constraint Code · Non-Negotiable Standards

Correctness is a product feature, not a backend preference.

The database schema, API contracts, and UI flows all carry the burden of truth — not just the last layer to touch the request.

Silent failures are design failures.

Retries, dead-letter handling, structured logs, and metrics exist before the first incident, not after the apology.

Every critical decision deserves a written rationale.

The next engineer should be able to understand the choice, the tradeoff, and what was consciously rejected.

If a system cannot be observed, it cannot be trusted.

Health checks, traces, and dashboards are treated as part of the product surface — not optional operations decoration.

Traced to live production decisions across TaxBridge · SabiScore · SwarmXQ

Lagos-built. Running globally.

Architecture calls documented →

Writing

Writing that ships decisions.

How the systems are actually built — architecture, ML decisions, and the production calls behind them. From Lagos to the world.

Start here if you want to understand how these systems were designed.

TIER 1STAFF+ SIGNAL3 min read

Building in Lagos, Running Globally: Infrastructure That Works Under Constraint

Why building production systems from Lagos — with unreliable power, variable network, and strict regulatory requirements — produces more resilient infrastructure than systems built in ideal conditions. Constraint is a design tool.

10 Jan 2026ARCHITECTUREPRODUCTION SYSTEMSRELIABILITY

Read the article

08 Dec 2025

SwarmXQ: How Self-Improving Agents Replace Manual Configuration at Scale08 Dec 2025

4 min

12 Nov 2025

Building SabiScore: Architecture Decisions in a Resource-Constrained Production System12 Nov 2025

3 min

15 Oct 2025

TaxBridge: Compressing 4 Hours of Nigerian SME Tax Filing to 15 Minutes15 Oct 2025

4 min

20 Sept 2025

Production ML Reliability: What 99.9% Uptime Actually Requires20 Sept 2025

4 min

05 Aug 2025

Zero-Downtime Design: The Baseline Every Production System Needs05 Aug 2025

4 min

See all 6 articles

3 systems. 4 certifications. Federal scale. Bring your constraint →

3 live production systemsFederal-scale delivery · 40M students99.9%+ sustained uptimeNRS · NDPC compliantLagos → Global

Contact

The system is ready. Are you?

Hiring for a senior engineering role, building from scratch, or containing a production incident — send it over. I respond within 24 hours, usually faster.

Shipped in LagosRunning globallyBattle-tested in audit seasonNRS · NDPC Compliant

STAFF+ / PRINCIPAL

Full-stack delivery · mobile app through production API

Staff+ and Principal roles at fintech and AI-native companies. Ownership of the entire surface — React Native mobile, Next.js dashboard, Fastify API, PostgreSQL RLS data layer. Multi-tenant isolation and zero-downtime deployments are baseline, not negotiated features.

TypeScript 5 · Effect-TS · React Native Expo 54 · Next.js 15 · Spring Boot · FastAPI · Turborepo.

TECHNICAL CO-FOUNDER

Pre-seed to Series A · Africa / emerging markets

Four years shipping production platforms from scratch under compliance pressure — NRS/DigiTax 2026, NDPC, and FIRS audit requirements. Systems that hold through due diligence, not just through the demo.

The system should outlast the seed deck. Available for full-time equity engagements.

INFRASTRUCTURE CONSULTING

Production reliability · compliance remediation · ML backends

Deliverable-led, not hourly. Scoped engagements: incident remediation, architecture review, Nigerian tax compliance (NRS 2026 / FIRS DigiTax), ML inference optimisation, and multi-tenant PostgreSQL RLS implementation.

You get working infrastructure and documented decisions — not a billable-hour report.

GitHub LinkedIn oscar@scardubu.dev Resume

Systems that work at 2am.
That's the standard.

Thesystemhastoworkat2am.

Thesystemhastoworkat2am.

Built to survive real constraints.

TaxBridge

SabiScore

SwarmXQ

Hashablanca

UBEC Data Pipeline

Shipped systems. Verified outcomes.

Infrastructure for problems nobody packaged yet.

pg-tenant

audit-chain

node-debug-llm

llm-dispatch

The stack behind systems that hold.

Federal scale. Production ML. Lagos → Global.

Writing that ships decisions.

Building in Lagos, Running Globally: Infrastructure That Works Under Constraint

The system is ready. Are you?

Built to survive real constraints.

TaxBridge

SabiScore

SwarmXQ

Hashablanca

UBEC Data Pipeline

Shipped systems. Verified outcomes.

Infrastructure for problems nobody packaged yet.

pg-tenant

audit-chain

node-debug-llm

llm-dispatch

The stack behind systems that hold.

Federal scale. Production ML. Lagos → Global.

Writing that ships decisions.

Building in Lagos, Running Globally: Infrastructure That Works Under Constraint

The system is ready. Are you?

Built to survive
real constraints.

Shipped systems.
Verified outcomes.

Federal scale.
Production ML.
Lagos → Global.

Built to survive
real constraints.

Shipped systems.
Verified outcomes.

Federal scale.
Production ML.
Lagos → Global.