PDavid Tandoh, Senior Software Engineer
I build AI platforms and agentic systems that are safe to run in production.
Senior software engineer with eight years in financial services. I design AI platforms, agentic systems and fraud analytics that hold up under real traffic and regulation. Founder of GuideX, a travel marketplace for Africa. Working remotely across the UK and Africa.
01The human
Eight years where mistakes cost money.
I’m David, a senior software engineer who has spent eight years building systems where mistakes cost money: payment switches in Ghana, fraud platforms for a UK bank, and now enterprise AI.
Recently I architected an internal AI platform that gives 50+ engineers secure access to 22 models through one gateway, with cost controls, guardrails and full tracing. I built the developer tooling that standardises how teams code with AI agents, and delivered fraud analytics that cut false positives by 35%.
Outside work I founded GuideX, an AI-enabled travel marketplace for African tourism, live on iOS and Android. I also run my own fleet of coding agents across two machines, and I write about how it works.
02Principles
Three rules I build by.
Resilient by default
Money-moving systems cannot fail quietly. I design for failure, observability and recovery first.
Intelligence with receipts
AI should explain itself. I favour signals and decisions a human can audit, never a model marking its own homework.
Lift others up
I mentor young people in tech with Kocha Mentors CIC and helped establish an offshore engineering team in Ghana.
On the desk right now
- Running an enterprise AI platform that gives 50+ engineers one secure gateway to 22 LLMs.
- Scaling GuideX, a travel marketplace for African tourism, live on iOS and Android.
- Running Firstmate as the chief-of-staff layer for a crew of coding agents after retiring Bosun.
- Shipping LLM-assisted fraud analytics for a UK bank.
- Mentoring underrepresented young people in tech with Kocha Mentors CIC.
03Work with me
Bring me the system that has to hold.
I take on work designing and building AI platforms, agentic systems and risk analytics, especially where reliability, observability and governance matter. Remote, across the UK and Africa.
Remote · UK & Africa
Agentic systems, end to end
From a first agent to a governed crew: orchestration, tool access, human approval gates and the evals that tell you whether it works.
- Agent and workflow design with context and harness engineering
- Model Context Protocol (MCP) servers that expose your services to agents safely
- Approval gates, guardrails and cost controls
Standardised AI-assisted development across 15+ repositories and deployed authenticated MCP infrastructure on AWS ECS.
Read the AI developer experience case studyAI and ML platforms
A secure, observable platform so your teams build on LLMs without each one managing providers, keys and spend.
- LLM gateways with multi-tenant governance and self-service provisioning
- Tracing, budgets and spend visibility from day one
- Infrastructure as code on AWS, Azure or GCP
Built a gateway to 22 enterprise LLMs used by 50+ engineers, deployed on AWS with Terraform.
Databricks Certified: Data Engineer Associate and Generative AI Engineer Associate.
Read the enterprise AI platform case studyFraud and risk analytics
Detection systems that hold up inside banking regulation and give investigators answers they can act on.
- Signal-driven risk pipelines with LLM-assisted classification
- ML infrastructure that data science teams can ship through
- Investigator tooling that joins data from many systems
35% fewer false positives on an ML-driven fraud detection platform for a UK bank.
Read the fraud analytics case study
Three ways to start
- Intro call
- You have a problem and want a straight read on it.
- A short conversation about the system, the constraint and whether I am the right person.
- Scoped build
- You know the outcome and need it designed and shipped.
- A fixed scope with a written plan, working software at each milestone and a documented handover.
- Embedded lead
- Your team needs senior hands on a platform or agentic programme.
- I join the team part time, set the architecture and standards, and build alongside your engineers.
Book a call
Choose a call, then a time that suits you. Tell me about the system, the constraint and what good looks like.
04.1Case study
Enterprise AI platform
- 22
- 50+
Problem
Every team that wanted to build with LLMs had to integrate providers, manage credentials and watch its own spend. That slowed delivery and left governance to chance.
Approach
- Architected a secure gateway that serves 22 enterprise LLMs, including Claude, GPT, Llama and Qwen, through one unified API.
- Deployed on AWS with LiteLLM, ECS Fargate, Bedrock, Terraform, ALB and WAF, with multi-tenant governance, self-service provisioning and enterprise guardrails.
- Built a developer portal in React, TypeScript, FastAPI and Cognito for self-service access, model discovery and credential management.
- Added MLflow tracing for end-to-end visibility into prompts, latency, cost and model behaviour.
Outcome
- 50+ engineers build AI-powered products without managing provider integrations.
- Organisation-wide budgets are enforced and spend is visible.
AWSLiteLLMECS FargateBedrockTerraformWAFMLflowReactFastAPICognito
04.2Case study
AI developer experience
- 15+
- MCP
Problem
AI coding assistants arrived team by team, each with its own prompts, habits and risks. Output quality depended on who was driving.
Approach
- Designed an internal AI developer platform that standardises AI-assisted development for Claude Code, Cursor, Kiro and GitHub Copilot.
- Built a reusable ecosystem of agentic workflows, context engineering, versioned AI assets and guardrails.
- Deployed secure Model Context Protocol (MCP) infrastructure on AWS ECS, exposing enterprise services and knowledge bases to agents through authenticated interfaces.
Outcome
- One standard for AI-assisted development across 15+ repositories.
- Consistent code generation and less onboarding effort for new engineers.
Claude CodeCursorKiroGitHub CopilotMCPAWS ECS
04.3Case study
Fraud analytics
- 35%
- £5m
- 50%
Problem
Fraud teams were working through too many false alarms, and new scam types such as invoice redirection needed detection that fits strict UK banking regulation.
Approach
- Built and automated the core infrastructure for an ML-driven fraud detection system with the data science team.
- Led the architecture and implementation of an invoice redirection scam detection application.
- Designed a signal-driven risk pipeline that aggregates indicators and applies LLM-assisted classification for investigation workflows.
- Contributed to an internal fraud agent tooling platform that joins data from multiple systems for case investigation.
Outcome
- False positives fell by 35%, with a forecast business impact of £2.5m a year.
- Invoice redirection detection carries a projected saving of £5m a year.
- Investigator call-handling time fell by up to 50%.
- Production data pipeline failures fell by 35% through proactive monitoring and self-healing.
ScalaSparkKafkaSnowflakeAWSTerraformGitLab CI/CD
Problem
Travellers who want to experience Africa struggle to find and trust local hosts, and hosts have no simple way to take bookings and get paid.
Approach
- Designed and built the mobile app, backend platform, booking infrastructure and cloud architecture, leading product strategy from concept to production.
- Shipped native iOS and Android apps for discovery, booking, in-app chat, QR-code tickets and payments.
- Built a two-sided marketplace with host identity checks through SmileID and payments in Ghanaian cedis through Paystack.
- Built an AI site-reliability agent that investigates log errors from the app and cloud infrastructure, with a human approving each case before a coding agent picks it up.
Outcome
- Live on the App Store, Google Play and the web, starting with hosts in Ghana.
- Discovery, booking, payment and host verification run on one platform.
FlutterNext.jsTypeScriptKotlinSpring BootFirebaseGoogle CloudPaystackSmileIDLangGraph
Problem
Running a dozen coding agents means a dozen terminals to watch, and one careless command can force-push, deploy or delete.
Approach
- A deterministic 24/7 outer loop watches agents and dispatches ready work in plain code, so autonomy stays in the inner loops.
- A tiered, fail-closed approval gate: routine work runs free, while force-push, deploy, rm -r and secrets pause for a tap on the phone.
- Per-project Kanban boards hold durable state, so a crashed orchestrator rebuilds from card status.
- Live spend telemetry, a cheaper default model and a hard daily cap keep cost bounded.
Outcome
- One conversation replaces a wall of agent terminals.
- Risky commands cannot run without a human decision.
- Firstmate superseded Bosun as the chief-of-staff layer.
PythonOmnigentHermesLLM agentslaunchdKanbanCEL
04.6On the bench
Also in the workshop.
DXJ Signal Engine
Building
A weekly research-not-advice signal for high-growth US stocks. A deterministic score ranks names; a deep-research agent writes cited bull and bear dossiers, and a CI firewall stops the LLM from ever touching the score.
PythonFastAPINext.jsPostgreSQLGitHub Actions
Cygnal
Building
Real-time fraud scoring for African banks and fintechs over ISO 8583 and REST, with deterministic rules against a sub-25ms target and a hash-chained audit trail built for central-bank examiners.
KotlinSpring BootWebFluxPostgreSQLRedisISO 8583
Cynthia
Prototype
A governed workflow that runs Claude as planner and reviewer and Codex as the only writer, with human approval gates and external completion checks instead of model self-report.
PythonOmnigentMCPClaudeCodex
05The journey
From card switches in Accra to AI platforms in the UK.
Each role added a layer: payment rails, then regulated data platforms, then the AI systems that run on both.
Senior Engineer
bigspark, for a UK bank, Nottingham, UK
Enterprise AI platform, AI developer experience, fraud analytics and platform leadership for a UK bank.
- Architected a secure gateway to 22 enterprise LLMs, used by 50+ engineers.
- Standardised AI-assisted development across 15+ repositories and deployed MCP infrastructure on AWS ECS.
- Built ML-driven fraud detection infrastructure that cut false positives by 35%, with a forecast impact of £2.5m a year.
- Led an invoice redirection scam detection application with a projected saving of £5m a year.
- Established and mentored an offshore engineering team in Ghana.
Why it matters. Regulated finance, cloud platforms and applied AI in one role: the blend behind banking’s move to intelligent, automated risk.
MSc, Software Engineering for Financial Services
University of Leicester, Leicester, UK
Master’s degree focused on engineering for financial systems.
- Awarded Best MSc Student in Software Engineering for Financial Services, 2020.
Why it matters. Formal grounding in financial systems, the foundation under the DXJ Signal Engine.
Card Systems Engineer
EProcess International (Ecobank), Ridge, Accra, Ghana
Owned critical card-payment systems and scheme integrations for a pan-African bank.
- Managed and supported Postilion Realtime and Postilion Office card systems.
- Led projects integrating National Payment Switches with the bank’s core payment switch.
- Technical liaison for international card schemes, including VISA and Mastercard.
- Redesigned incident management and escalation, cutting incident response time by 50%.
Why it matters. Deep payments-rails experience with ISO 8583, switching and card schemes: the operational reality behind modern fintech.
EFT Software Engineer
Electronic Funds Technology Corporation, Labone, Accra, Ghana
Payment integrations for card processing across multiple African markets.
- Developed Java-based payment integrations and reconciliation tooling on the Postilion Framework.
Why it matters. Where it started: the discipline of money-moving systems that cannot fail.
BSc, Computer Science
Ashesi University, Berekuso, Ghana
Computer Science degree from one of Africa’s leading universities.
- Foundations in software engineering, algorithms and systems.
Why it matters. Rigorous fundamentals that scale from card switches to AI agents.
06Capabilities
The toolkit, by layer.
- AI platforms and LLM engineeringDatabricks Certified Generative AI Engineer Associate
- LiteLLM gatewaysBedrockOpenAIAnthropicMCPRAGVector databasesEmbeddingsTool callingMLflowEvalsContext engineeringHarness engineering
- Agentic engineering
- CrewAILangChainLangGraphClaude CodeCursorCodexKiro
- Cloud and DevOps
- AWSAzureGCPDockerKubernetesTerraformGitHub ActionsGitLab CI/CD
- Data engineeringDatabricks Certified Data Engineer Associate
- DatabricksSparkSnowflakeKafkaAirflowPostgreSQLMongoDB
- Languages and frameworks
- JavaKotlinSpring BootPythonFlaskScalaGoC#ReactRESTGraphQL
- Fintech and payments
- ISO 8583ISO 20022Payment switchesFraud detectionPCI DSSPostilion
- Quality and practice
- TDD and BDDJUnitMockitoJestKotestTestcontainersAgile
07Credentials
Certified, and checkable.
Certified by Databricks in data engineering and generative AI engineering, and by Anthropic and Microsoft. Every card opens the issuer’s public record.
Certified Generative AI Engineer AssociateGenerative AIDatabricksVector SearchModel ServingMLflowUnity CatalogRAG applicationsLLM chainsVerify with Databricks: Databricks Certified Generative AI Engineer Associate
Certified Data Engineer AssociateApache SparkDelta LakeDatabricksLakehouseDelta Live TablesData PipelinesETLProductionVerify with Databricks: Databricks Certified Data Engineer Associate
Claude Certified Architect - FoundationsVerify on Credly: Claude Certified Architect - Foundations
StreamSets White BeltVerify on Credly: StreamSets White Belt
Cloud platforms
AWS
Where the enterprise AI platform runs: ECS Fargate, Bedrock, ALB and WAF, provisioned with Terraform.
Azure
Part of the cloud toolkit, backed by the Microsoft fundamentals certification.
Google Cloud
Where GuideX runs: Google Cloud and Firebase behind the live apps.
Platform
Awards
Building Visionary Partnerships Best MSc Student in Software Engineering for Financial Services
Verified credentials. Each issuer confirms its own record through the links above. The full list is also on my LinkedIn profile and my Credly profile.
08Let’s build
Have a system that has to hold? Let’s talk.
I take on contracts in AI platforms, agentic systems and risk analytics. The quickest route is a short call.