At a glance
The system we operate (this site)
1,583
catalog rows shipped
live production data
71
unit tests
regression-pinned to incidents
$61
monthly GCP spend
4 services, fully managed
−41%
cost vs prior month
BigQuery-driven optimization
89+
PRs merged
this quarter, main branch
10
cron-driven generations/day
zero gate-failure publications
4
LLM providers wired
Grok, Gemini, Claude, Bedrock
0
downtime incidents
last 90 days
Every catalog row carries source URL, raw quotes, factor hash, and pain signal strength so any claim is auditable end-to-end. Full methodology disclosure available on request during engagement.
System case study
StartupTribunal — engineering depth
What a CTO actually wants to see — the security, reliability, eval, cost, and deployment posture of a system in production, as the team runs it day to day.
Architecture
Next.js 16 on Vercel Fluid Compute (Node 24), 3 GCP Cloud Run workers (auto-catalog-generator, tribunal-worker, mcp-toolbox), Cloud SQL Postgres 15 via Cloud SQL Connector with Workload Identity Federation, GCS blob offload for large JSON columns (8 columns per row), Cloud Tasks for async pipeline hand-offs, LangGraph 1.x StateGraph for the country-relevance gate, pluggable multi-LLM router across Grok 4, Gemini, Claude, with AWS Bedrock-backed VibeJudge for code judging.
Security
Workload Identity Federation for GCP service-to-service auth (no JSON keys in CI). Firebase Admin SDK for user auth; x-admin-key for internal admin routes; separate x-cron-secret for Cloud Scheduler so cron jobs cannot escalate to admin. Secrets live in GCP Secret Manager (8 secrets), never in committed env files. Pre-commit gitleaks scan blocks accidental credential commits. CI runs npm-audit at --audit-level=critical. Stripe / Paystack payment intents server-side only. PII redacted at the Sentry boundary.
Reliability & Observability
Health endpoint with three modes: /api/health, ?minimal=true (liveness), ?deps=true (DB + GCS connectivity, returns 503 on failure). 10 Cloud Scheduler jobs covering 24-hour catalog rotation, each idempotent. Slot-skip on gate failure — pipeline refuses to publish low-confidence rows under degradation. Per-row provenance preserves source_url, raw_quotes[], factor_hash, and pain_signal_strength so any claim is re-traceable. Sentry on both Vercel and Cloud Run with structured logging.
Eval discipline & anti-hallucination
Validation retry loop: schema check (non-retryable), anchor enforcement (mandatory ≥1 code anchor on learn/ship days), URL liveness HEAD-probe with GET fallback for arXiv-style endpoints — up to 2 quality-retries with prior issues fed back into the prompt. Country-relevance gate runs 4 layered rules (target-in-title accept, competing-country reject, 3:1 body dominance ratio, snippet fallback). Title sanitizer strips Grok rationale leaks and markdown wrapping. Every gate rule has a regression test pinning an exact production incident. The team has a promptfoo-based eval harness on the roadmap.
Cost engineering
BigQuery billing export wired from day one for per-SKU forensics. Identified GCS egress as the dominant cost driver via SKU breakdown, shipped a 24-hour blob cache using Next 16 unstable_cache with revalidateTag invalidation. Result: −57% Cloud Storage, −41% total monthly bill. Artifact Registry cleanup policy (delete untagged after 7d, keep tagged). Cloud Run revision prune (47→5 per service). Cloud SQL PITR retention tuned. The full audit and the resulting deltas are documented in CLAUDE.md.
Deployment discipline
Our standard ship cycle: audit → refactor for testability → TDD (RED → GREEN) → typecheck → eslint → build → PR → CI → squash merge → deploy → live verification. Vercel deploys via ./scripts/deploy.sh which wraps vercel --prod with a post-deploy health gate that curls 5 critical pages and fails loudly on any 5xx. Cloud Run deploys via path-filtered GitHub Actions (only the changed service rebuilds). Conventional Commits + release-please for automated CHANGELOG and semver. No direct-to-main in normal operation — feature branches and PRs are the default.
How I think
Engineering principles
- 01
TDD-first on every gate rule.
Every production incident becomes a regression test that pins the exact failing example. The 4-rule country gate has 4 distinct test files, each tied to a real prod row.
- 02
Honest fallbacks over silent degradation.
If a pipeline can't ship a high-confidence row, it skips the slot rather than publishing a Frankenstein one. The user sees fewer rows, not bad rows.
- 03
Provenance as a first-class concern.
Every catalog row carries source URL, extracted raw quotes, content-hash, and signal strength so any claim can be audited end-to-end.
- 04
Cost as a feature, not an afterthought.
BigQuery billing export was wired from day one. The team identified the dominant SKU, shipped a targeted fix, and measured the delta in a week. −41% spend in 30 days.
- 05
Documentation as code.
Internal runbooks, public-facing system disclosures, and per-component docs — all version-controlled, all walked through in PRs.
- 06
Africa-built, global-grade.
Same engineering standards as a SF SaaS, designed under the constraints that matter for emerging markets — latency, cost, connectivity, fallback paths. These constraints are showing up in US enterprise too.
Engagement models
What we build for clients
Three sharply scoped packages. Each lists what ships and what doesn't — the absence is part of the contract.
AI product MVP
4–8 weeks · fixed scope + budget
From idea to deployed first version. You get a production codebase, CI/CD, evals, methodology disclosure, and a 30-day handoff window — the same standard we hold our own systems to.
- ✓Production codebase + CI/CD
- ✓Eval harness on the AI components
- ✓Public methodology disclosure (recruiter / partner audit trail)
- ✓Cost model + runbook
- ✓Knowledge transfer + 30-day handoff
- ·Visual design / branding
- ·Marketing / GTM
- ·Equity-only deals
AI feature integration
2–4 weeks · per-feature pricing
RAG, agents, multi-LLM routing, evals, or fine-tunes added to an existing application. Production-grade or we don't ship it.
- ✓Production integration in your codebase
- ✓Eval suite for the new behavior
- ✓Cost model with sensitivity analysis
- ✓On-call handoff doc + 14 days of post-ship support
- ·Greenfield model research
- ·Multi-vendor migration projects
- ·Replatforming your existing infra
Embedded FDE / contract IC
1–3 months · senior engineer rate
Forward Deployed Engineer-style embedding with your team. Daily PRs, weekly demos, clean handoff when the engagement ends. Lead engineer is Maku; the team supports.
- ✓Daily Slack/email updates
- ✓PR-driven work with reviews
- ✓Weekly demo + scope check
- ✓Documentation of every system touched
- ✓Clean handoff package at exit
- ·Sales / BD / partnerships work
- ·Legal / compliance review
- ·Pure design without engineering
Shipped work
Selected projects
StartupTribunal
Founder & Lead Engineer · Dec 2024 → present
Problem: No fast way to surface validated, country-tagged pain signals for African and emerging-market builders.
Solution: Multi-LLM ranking + 4-rule country gate + structured pain extraction with full provenance. 1,583 catalog rows live, 71 unit tests pin every regression, $61/mo infra. Hybrid Vercel + GCP Cloud Run, 3 microservices, Cloud Tasks for async work.
Outcome: Production system serving live traffic. 89+ PRs merged this quarter, 41% MoM cost reduction shipped via BigQuery-driven analysis. SEO infrastructure scored 97/100 across 150+ programmatic pages, dynamic sitemap with 1,000+ URLs.
Next.js 16 · TypeScript · LangGraph · GCP · Cloud SQL · Grok 4 + Gemini + Claude · AWS Bedrock
startuptribunal.com →VibeJudge
Founder & Engineer · 2024 → present
Problem: Hackathon submissions need fast, defensible code review — humans don't scale, and naive LLM scoring is uncalibrated.
Solution: AWS Lambda-backed multi-agent system. 5 Bedrock agents (bug_hunter, innovation, performance, code_quality, ai_detection), each Pydantic-validated. Weighted score aggregation with file:line bug citations. Repository never retained — clone, analyze, discard.
Outcome: Live judging service integrated with StartupTribunal hackathons. Documented Pydantic-validation failure modes (e.g. evidence list length cap) and shipped fixes — postmortem published in CLAUDE.md.
AWS Lambda · SAM · Bedrock (Nova Lite) · Python · Pydantic · DynamoDB
AgriGuard farm advisory
Technical Lead · Dec 2023 → Nov 2024
Problem: Smallholder farmers needed AI-assisted advisory grounded in local agronomic knowledge, plus an insurance platform that could handle complex business logic and geospatial data.
Solution: Django + Celery + Docker backend. RESTful APIs with proper authentication. GeoDjango for spatial queries. RAG pipeline for the AI advisory system using LangGraph-style orchestration. PostgreSQL tuned with indexing, query optimization, connection pooling. CI/CD via GitHub Actions.
Outcome: Production deployment serving real users. 70% reduction in deployment time vs prior manual process. Integrated Stripe and Paystack payment processing with proper authentication. Led cross-functional team through full product lifecycle.
Django · DRF · Celery · PostgreSQL · GeoDjango · Docker · GCP · Stripe + Paystack
Farmerline mobile platform
Senior Mobile Developer · Oct 2024 → present
Problem: Mobile applications serving agricultural stakeholders across Africa — performance, reliability, and clean-architecture rigor at scale.
Solution: Kotlin / Jetpack Compose. Cross-functional Agile sprint delivery. Code reviews and technical documentation discipline. Working on production releases serving real farmers.
Outcome: Active role; references available on request.
Kotlin · Jetpack Compose · Android · Agile
Track record
Experience
Senior Mobile Developer · Farmerline Group
Oct 2024 — Present
Accra, Ghana
- ·Develop and maintain mobile applications serving agricultural stakeholders across Africa.
- ·Collaborate with cross-functional teams in Agile sprints to deliver features on schedule.
- ·Implement scalable solutions following software engineering best practices.
- ·Contribute to code reviews and technical documentation.
Founder & Lead Engineer · Genesis Protocol — StartupTribunal
Dec 2024 — Present
Accra, Ghana
- ·AI-powered startup validation marketplace; 1,583 catalog rows in production on hybrid Vercel + GCP architecture.
- ·Built 4-rule country-relevance gate + multi-LLM router; shipped 89+ PRs this quarter.
- ·Designed cost-optimized infrastructure on ~$61/mo via BigQuery-driven optimization; −41% MoM.
- ·Integrated 4 payment providers (Stripe, Paystack, Flutterwave, PayPal) with race-condition prevention.
- ·Cloud Tasks async job processing, Cloud SQL Connector with retry + backoff, Redis-backed view counter.
Technical Lead · AgriGuard Ltd
Dec 2023 — Nov 2024
Accra, Ghana
- ·Architected backend using Django, Celery, Docker for agricultural insurance platform.
- ·Designed RESTful APIs handling complex business logic and geospatial data.
- ·Established CI/CD pipeline via GitHub Actions; reduced deployment time by 70%.
- ·Integrated secure Stripe and Paystack payment processing with authentication.
- ·Led AI-powered farm advisory using RAG pipeline orchestration.
- ·Optimized PostgreSQL with indexing, query optimization, connection pooling.
- ·Led cross-functional team through full product development lifecycle.
Entrepreneur In Training · MEST Africa
Sep 2023 — Aug 2024
Accra, Ghana
- ·Completed 12-month intensive entrepreneurship and technology training program.
- ·Built full-stack applications with focus on scalability, clean architecture, maintainability.
- ·Gained deep experience in business strategy, product development, technical leadership.
Android Engineer · Accenture
Jun 2022 — Apr 2023
Nairobi, Kenya
- ·Developed enterprise-grade Android applications for financial services clients.
- ·Collaborated in Agile teams using SCRUM methodology with 2-week sprints.
- ·Implemented authentication mechanisms and secure API integrations.
- ·Wrote comprehensive unit and integration tests; conducted code reviews.
Full-Stack Developer · Pelel Enterprises
Apr 2021 — Mar 2022
Arua, Uganda
- ·Built farmer information management system: Android frontend, PHP Lumen backend.
- ·Designed RESTful APIs with proper versioning for mobile-backend communication.
- ·Implemented PostgreSQL schema for agricultural loan processing and reporting.
- ·Deployed backend with CI/CD pipeline; integrated JWT-based auth.
Android & Web Developer · Freelance & Contract
2020 — 2021
Remote · Kenya / Uganda / Ghana
- ·Delivered full-stack solutions using Vue.js, Firebase, modern frameworks.
- ·Built admin panels and e-commerce features with secure auth and payments.
- ·Collaborated remotely with distributed teams across Africa.
Foundation
Education & certifications
Bachelor of Information Technology · Uganda Christian University
2015 – 2018
Android Development Certification · Moringa School
2019
Entrepreneurship & Technology Training Program · MEST Africa
2023 – 2024
Certifications
- · Google Africa Developer Scholarship — Forward Program
- · Enterprise Design Thinking Practitioner (IBM)
- · Agripreneurship: A Path to the Future
Technical expertise
Stack & skills
Languages
TypeScript (10y), Python, Kotlin, Java, Go (working knowledge), SQL, Bash
AI / ML
OpenAI, Anthropic, xAI Grok, Google Gemini, AWS Bedrock; LangGraph 1.x; RAG pipelines; prompt-versioning; eval design; multi-LLM routing; cost-per-token economics
Cloud & infra
GCP (Cloud Run, Cloud SQL, GCS, Cloud Tasks, Cloud Scheduler, Secret Manager, Workload Identity Federation, BigQuery, Cloud Build); AWS (Lambda, Bedrock, SAM, DynamoDB); Vercel (Fluid Compute); Firebase; Railway; DigitalOcean
Frameworks
Next.js 16 (App Router), React 19, Django + DRF, FastAPI, PHP Lumen, Jetpack Compose, Flutter, Vue.js, Tailwind CSS
Data
PostgreSQL 15 (with pgvector exposure), JSONB columns at scale, GeoDjango (spatial queries), Cloud SQL Connector with retry + backoff, BigQuery analytics, Firebase Firestore, Redis
Architecture
REST API design, microservices, RAG pipelines, async job processing (Cloud Tasks, Celery), authentication systems (OAuth 2.0, JWT, session, Firebase Admin), webhook handling with idempotency
Testing & quality
Jest, RTL, JUnit, integration tests, TDD discipline, test-pinning of production incidents, code review culture, OpenAPI documentation
Security
Workload Identity Federation for keyless service auth, secret rotation in GCP Secret Manager, rate-limiting at API boundary, gitleaks pre-commit, npm-audit CI gates, PII redaction at observability boundary, server-side-only payment intents
Observability
Sentry (Vercel + Cloud Run), structured logging, BigQuery billing export for cost forensics, per-row provenance tracking, admin audit logs, health-endpoint with deps probe
DevOps
Docker multi-stage builds, GitHub Actions (with path filters), GitLab CI, Cloud Build, conventional-commits + release-please for automated semver + CHANGELOG, blue/green deploys, post-deploy health gates
Mobile
Android (Kotlin, Java), Jetpack Compose, Flutter (cross-platform), Firebase integration, secure API consumption from mobile, offline-first patterns
AI tooling
Claude Code (this site largely built with it), Figma MCP (Android design workflows), Atlassian MCP (QA issue tracking)
Methodology
Agile / Scrum, Clean Code principles, SOLID, code reviews as a discipline, OpenAPI-first API design, technical mentorship of junior engineers, technical interviewing
Working style
How we work with clients
Async-first, predictable cadence
Daily written updates over Slack or email. Weekly demos with running notes. You always know where the work is.
PR-driven
Every change reviewed before it merges. No direct-to-main in normal operation. Conventional commits + CHANGELOG so handoffs survive the engagement.
Methodology page per project
We write a public methodology disclosure for every shipped system — recruiter and partner audit trail, same level of rigor we hold our own work to.
Honest scoping
If your project is wrong-shape for us, we'll tell you and refer. We'd rather pass than ship a half-finished system you have to inherit.
Clean handoff
At the end, your team owns the system. Tested, documented, runnable without us. 30-day post-ship support window for MVPs is standard.
Real timezones
Based in Accra (GMT). Overlap with EU mornings and US afternoons. Comfortable in distributed teams across Africa, EU, and US.
Three ways to start
Get in touch
Book a call
30 minutes, free, no agenda needed. Best for recruiters, partner intros, and quick fit checks.
Pick a time →Send a brief
Five quick questions. I'll read it before we talk so the call is focused and you don't pay for me to catch up.
Start brief ↓Just say hi
Email or LinkedIn. Best for one-off questions, mentorship pings, and general connections.
makpalyy@gmail.com →LinkedIn →Build a brief
Tell me about your project
Five questions, ~2 minutes. Lands in my inbox. I reply within 48h, or skip the line and book a call directly.