Open to Applied AI Engineer roles

I build production
AI systems.

Applied AI engineer specializing in LLM pipelines. I build retrieval, agent orchestration, evals, and the backends that keep them reliable in production.

Get in touchResume ↓
Full-Stack + AI
Live Product
Open Source Work
Featured Project
Live · Streaming CLI

fourpoket

An AI coding agent built on multi-provider orchestration with retries and fallback, AST-aware code operations, and per-call cost tracking.

Built withTypeScript · Node.js · Next.js · Stripe
fourpoket.com ↗npm: fourpoket ↗Full deep dive ↓

Experience

Production systems designed and built inside companies, with a focus on applied AI.

Applied AI / LLM systems
Etiqa InsuranceInsuranceApr 2024 - Nov 2025

Claims Operations Assistant

Senior Applied AI Engineer

An internal assistant for a claims operations team: hybrid retrieval with reranking, drafting behind a code-enforced guardrail gate, and a supervisor/worker agent layer with structured outputs. A routing classifier sends each query to an off-the-shelf Q&A layer, the custom pipeline, or a direct structured lookup, scoped as augmentation with a human in the loop.

Agent layer + MCP serverHybrid retrieval + rerankingRAG-vs-structured-query routingLLM evals + regression gateGuardrail-enforced HITL boundaryCost-aware model routingMulti-tenancy (Postgres RLS)Observability
Core stackPython · FastAPI · Pydantic · pgvector · Postgres RLS · MCP · OpenTelemetry
Outcomes
Ran in production across multiple delivery phases as a daily tool for the operations team
Served knowledge Q&A and drafting on one shared retrieval pipeline
Retrieval precision improved across phases through hybrid search and reranking
Routing kept the bulk of traffic on a cheaper model tier without quality regression
Extended to a second business entity with isolated data and scoped evals
The augmentation boundary was enforced mechanically rather than by convention
FreelanceB2B SaaSApr 2023 - Mar 2024

AI Knowledge Chatbot Platform

Senior Applied AI Engineer

An upload-your-docs-get-a-chatbot product covering the full path from ingestion through hybrid retrieval to streaming RAG orchestration, with multi-model routing and semantic caching. The product layer shipped with it: admin dashboard, embeddable widget, integrations, and billing.

RAG pipelineHybrid retrieval + rerankingStreaming responsesMulti-model routingSemantic cachingIntegrations (Slack / Intercom / Notion)
Core stackTypeScript · Node.js · Python · FastAPI · pgvector · Redis
Outcomes
Went from prototype to a live product in steady use
Retrieval quality climbed across the engagement through hybrid search and reranking
Per-query cost dropped meaningfully from caching plus routing, with margins healthy
Integration depth (Slack, Intercom, Notion, Drive) differentiated against lighter competitors
Multi-tenant data held cleanly with no cross-tenant leakage and reconciled billing
Full-stack foundation
AccentureLogisticsJan 2022 - Mar 2023

Document Intelligence Platform

Senior Software Engineer

A document-intelligence product taken from early stage to production, on a two-service architecture: a Node and TypeScript product layer alongside a Python OCR and NLP extraction service, joined by a typed contract.

Multi-tenant SaaSTwo-service architectureKeyboard-first review UXTyped service contracts
Core stackNext.js · Node.js · TypeScript · Python · FastAPI · PostgreSQL
AccentureCorporate servicesNov 2020 - Dec 2021

Corporate Workflow & Approval System

Senior Software Engineer

A travel-and-expense workflow system rebuilt and shipped to production. The core design is a composable architecture where workflow templates and form templates are independent reusable building blocks configured as data, so new request types ship as configuration rather than code.

Team leadComposable workflow engineState machineAudit trail
Core stackReact · TypeScript · Node.js · PostgreSQL · Redis · SAML SSO
SunwayEvents securityNov 2019 - Oct 2020

Event Security Management System

Software Engineer

The backend API and web operations dashboard for an outdoor-events security operation, with API contracts coordinated with a separate mobile team. Replaced spreadsheet scheduling and paper attendance with a constraint-based scheduling engine, QR check-in with GPS verification, and a real-time operations dashboard.

Constraint-based schedulingReal-time dashboardSpatial queries (PostGIS)Cross-team API contracts
Core stackNode.js · TypeScript · PostgreSQL · PostGIS · Socket.IO · Redis

fourpoket

A complete look at the architecture, engineering decisions, and capabilities behind the product.

fourpoket.com npm install

Architecture

Four repositories, three deployment targets, one cohesive product.

four-poket-backend
Express API · AI orchestration · SQLite · multi-tenant
Railway
four-poket-cli
Ink terminal UI · AST parsing · code read/write engine
npm
four-poket-web
Next.js 16 · Auth · Dashboard · Stripe billing
Vercel
four-poket-admin
React + Vite · Session explorer · Revenue overview
Local
Domain via Cloudflare · Monitoring via Sentry · Emails via Resend

Tech Stack

Backend
TypeScriptNode.jsExpressSQLiteLevelDBZod
Frontend
ReactNext.js 16Tailwind CSS
CLI
Commander.jsInk (React)Web WASM
Auth & Payments
JWTGoogle OAuthStripebcrypt
AI Providers
DeepSeekMiniMaxOpenRouter
DevOps & Infra
RailwayVercelCloudflareDockerSentryResend
Admin Dashboard
ReactViteReact QueryRechartsReact Router
Testing
JestUnit TestsIntegration Tests

Engineering Highlights

The key technical achievements. Click any card to expand.

Open Source

Personal AI systems work, built in public with readable source.

Personal projectAI toolingGitHub ↗2026

learning-system

Public source, actively developed

A local learning platform where an LLM acts as the teacher and a typed Python backend owns memory, orchestration, and context engineering. The public source spans the applied AI stack: provider abstraction over two model transports, pgvector retrieval, a versioned eval framework with regression reporting, OpenTelemetry tracing, and an approval-gated agent layer.

Agent orchestrationEval frameworkLLM observabilitypgvector retrievalTool callingHuman-in-the-loop
Core stackPython · FastAPI · Pydantic · PostgreSQL · pgvector · OpenTelemetry · pytest
Outcomes
Public commit history with signed commits and pre-commit gates for lint, strict typing, and commit format
Versioned eval sets run against either transport with deterministic and LLM-as-judge scoring
A regression report diffs each eval run against the last, per set and per item
Every LLM round trip is recorded with latency and cost fields and linked to traces and error logs
Agent mutations apply atomically behind a human approval gate, and failures survive rollback
Smoke scripts verify transport contracts against live providers, separate from the unit suite

About

Hey, I'm Kent. I build the layer between raw model capability and software people can actually trust.

Here's the bet I'm making with my career. Every powerful tool has always charged an entry fee. Blender, Photoshop, a serious spreadsheet, each one takes months before it gives anything back. AI collapses that fee into a conversation. You say what you want and the tool meets you there. Wrapping hard software in plain language is the biggest shift in how people use computers since the GUI, and it gets won or lost at the application layer. That's where I work.

A model on its own is an engine on a stand. Loud, impressive, going nowhere. I build the car around it: retrieval and orchestration as the drivetrain, evals as the brakes, observability as the dashboard. I work the whole machine because the interesting failures hide between the parts. A retrieval bug can look like a prompt problem, and a cost spike can really be a chunking decision. Raw capability is getting cheap. Proof that it works is the hard part, and that's the part I love.

Applied AI
LLM pipelines, retrieval, agents, and evals.
Full-Stack
Backend, frontend, infra, and everything from embeddings to evals in between.
Live Product
A production AI tool, live and taking real payments.
End-to-End Ownership
From model behavior to billing webhooks, the whole system.
What I Bring
LLM OrchestrationRAG & Hybrid RetrievalAgent SystemsLLM EvaluationStructured OutputsStreamingHuman-in-the-Loop DesignContext EngineeringPrompt EngineeringCost & Latency Optimization

Get in Touch

Interested in working together? Let's connect.

Open to full-time Applied AI Engineer roles · Remote

hello@kentdoki.devGitHubLinkedInfourpoket.comnpm: fourpoket