Fredrick Onyango · Software EngineerDownload CV

AI & Engineering

AI systems, designed end to end

The pipelines I build and the architecture decisions behind them — from retrieval to deployment.

AI & Engineering

Beyond calling an AI API

AI features only work as products when the engineering around them does: retrieval quality, fallback chains, data models and security. This is the pipeline behind Kilele Archade's tutor.

kilele-archade / ai-tutor-request

  1. Student questionfrom a subject, level and competence
  2. Frontend (React + TypeScript)
  3. API layerSupabase
  4. Authentication & row-level security
  5. Application logicresolve curriculum scope for the query
  6. Document retrievalmetadata filter + pgvector similarity search
  7. Relevant contextre-ranked passages within token budget
  8. LLM / AI providerprimary provider with transparent fallback chain
  9. Grounded response + citationsCloudinary assets attached where relevant

What I work with

Retrieval-Augmented GenerationEmbeddingsVector databases / pgvectorDocument processingLLM integrationAI provider fallback architecturePrompt engineeringCloud AIMachine learningComputer visionAI system architecture

How I think about AI systems

  • Retrieval quality, not model size, decides whether answers are trustworthy.
  • Every provider is replaceable — build the fallback before you need it.
  • Ground answers in your own data and cite it, so users can verify.
  • An AI feature is a full pipeline: ingestion → indexing → retrieval → prompting → evaluation.

System Architecture

How the products are put together

Selected architectures from shipped work — authentication, business logic, data, external services and the AI layer, each in its proper place.

kilele-archade / system-overview

  1. Student / Teacher / Parent
  2. React + TypeScript frontend
  3. Supabase API & Auth (row-level security)
  4. Application logic & role checks
  5. Document retrieval — pgvector similarity search
  6. Relevant curriculum context assembled
  7. LLM provider with fallback chain
  8. Grounded response + citations

Kilele Archade: architecture, security and decisions →

apex / system-overview

  1. Owner / Manager / Tenant
  2. React frontend
  3. API layer with authentication
  4. Role-based authorisation
  5. Business logic — leases, ledger, maintenance
  6. PostgreSQL (properties, tenants, payments)
  7. External services — M-PESA, SMS gateway
  8. Reports & notifications

Apex: architecture, security and decisions →

mediflow / system-overview

  1. Reception / Clinician
  2. React frontend
  3. Express API with auth
  4. Clinic logic — scheduling, queue
  5. PostgreSQL

MediFlow: architecture, security and decisions →