Roadmap
What I am building next.
A twelve-month plan from October 2026. Each phase ends with running code and a written account of it.
- Oct — Nov 2026
AI features that are tested
Now
Evaluations for a retrieval (RAG) endpoint that run in CI and fail the build when answers get worse, next to the security scans already in the pipeline.
RAGEvaluationsGitHub ActionsOWASP ZAPOn the record when done
A write-up on how the endpoint is tested
- Dec 2026 — Feb 2027
Agents and the plumbing under them
Next
Hybrid search on PostgreSQL with pgvector, an agent loop written by hand and then in LangGraph, and an MCP server, with tracing and per-request cost limits.
PostgreSQLpgvectorLangGraphMCPTracingOn the record when done
Evaluation scores before and after
A write-up of the MCP server
- Feb — May 2027
A distributed platform, run like production
Later
Go and TypeScript services over Kafka, PostgreSQL, and Redis on Kubernetes, provisioned with Terraform and deployed with ArgoCD. OpenTelemetry, alerts, load tests, and the cost of each request.
GoKafkaKubernetesTerraformOpenTelemetryOn the record when done
Architecture decision records
A postmortem of a deliberate failure
AWS Certified Solutions Architect – Associate
- May — Jul 2027
Security for cloud and AI systems
Later
Prompt-injection tests against my own AI endpoints, guardrails in CI, OAuth and OIDC done properly, least-privilege IAM, and a signed, inventoried supply chain.
OWASP LLM Top 10OAuth / OIDCIAMSBOMThreat modelingOn the record when done
A prompt-injection write-up with the fixes
CompTIA Security+
- Jul — Oct 2027
An AI operations platform
Later
Agents that read logs, traces, and alerts from the platform above, investigate failures, and propose fixes. Anything risky waits for a person to approve it.
AgentsObservabilityHuman approvalEvaluationsOn the record when done
A case study in Work
A talk at a local meetup
Week by week
Four-week blocks from 19 October 2026. Each block has one book, one or two courses, something to build, and something to publish.
- Weeks 1–4
19 Oct – 15 Nov 20260 · Finish in-flight
- Read
- AI Engineering (Chip Huyen), chapters 1–4, focused on evaluation
- Courses
- DeepLearning.AI "Automated Testing for LLMOps"; finish NetAcad Networking Basics
- Build
- OWASP ZAP in CI; a golden set and evals for /ask in GitHub Actions
- Publish
- "How I test my RAG endpoint"; CV rewritten as problem-owner stories
- Weeks 5–8
16 Nov – 13 Dec 20261 · AI core
- Read
- AI Engineering, chapters 5–10
- Courses
- Anthropic courses (API fundamentals, tool use); DeepLearning.AI "Building and Evaluating Advanced RAG"
- Build
- /ask: hybrid search, a reranker, citations, and a move to PostgreSQL with pgvector
- Publish
- Evaluation scores before and after the /ask upgrade
- Weeks 9–12
14 Dec 2026 – 10 Jan 20271 · AI core and Go
- Read
- Understanding Distributed Systems (Roberto Vitillo)
- Courses
- Learn Go with Tests and the Exercism Go track; DeepLearning.AI "AI Agents in LangGraph"
- Build
- An agent loop written by hand, then in LangGraph; a first small Go service
- Publish
- "An agent loop from scratch"
- Weeks 13–16
11 Jan – 7 Feb 20271 · AI core and data
- Read
- Designing Data-Intensive Applications (Kleppmann), part I; Use The Index, Luke
- Courses
- DeepLearning.AI "MCP: Build Rich-Context AI Apps with Anthropic"; Hugging Face Agents Course
- Build
- An MCP server for my API; Langfuse tracing; cost and latency limits
- Publish
- A write-up of the MCP server
- Weeks 17–20
8 Feb – 7 Mar 20272 · Platform
- Read
- Designing Data-Intensive Applications, part II: replication, partitioning, transactions, consistency
- Courses
- Confluent "Kafka 101"; MIT 6.5840 Distributed Systems, lectures 1–6
- Build
- A distributed platform: Go and TypeScript services, Kafka, PostgreSQL, Redis
- Publish
- The first architecture decision records
- Weeks 21–24
8 Mar – 4 Apr 20272 · Platform
- Read
- Kubernetes: Up and Running; Terraform: Up & Running (selected chapters)
- Courses
- AWS Solutions Architect Associate course (Adrian Cantrill or Stephane Maarek)
- Build
- The platform on EKS or k3s with Terraform, Helm, and ArgoCD
- Publish
- An architecture diagram and an article on its trade-offs
- Weeks 25–28
5 Apr – 2 May 20272 · Platform
- Read
- Observability Engineering (Majors, Fong-Jones, Miranda); Google SRE book chapters on SLOs and postmortems
- Courses
- Linux Foundation LFS148 "Getting Started with OpenTelemetry"; Tutorials Dojo practice exams, then the AWS Solutions Architect Associate exam
- Build
- OpenTelemetry tracing, alerts, k6 load tests, a deliberate failure and its postmortem, cost per request, and a self-hosted model compared with the API
- Publish
- The postmortem and a cost write-up
- Weeks 29–32
3 May – 30 May 20273 · Security
- Read
- Threat Modeling: Designing for Security (Adam Shostack), selected; OWASP Top 10 for LLM Applications
- Courses
- PortSwigger Web Security Academy (core labs); Lakera Gandalf prompt-injection game
- Build
- Attacks on my own /ask endpoint and MCP server; guardrails and injection tests in CI
- Publish
- A threat model of the platform
- Weeks 33–36
31 May – 27 Jun 20273 · Security
- Read
- API Security in Action (Neil Madden): OAuth 2.0, OIDC, JWT
- Courses
- Professor Messer Security+ videos
- Build
- OIDC sign-in on the platform; least-privilege IAM; secrets; Kubernetes network policies
- Publish
- "OAuth mistakes I fixed in my own API"
- Weeks 37–40
28 Jun – 25 Jul 20273 · Security
- Read
- Hacking Kubernetes (Martin, Hausenblas), selected; the SLSA docs
- Courses
- Security+ practice exams, then the Security+ exam; TryHackMe cloud security rooms
- Build
- SBOMs, pinned dependencies, and signed images (cosign) in CI
- Publish
- A prompt-injection write-up with the fixes
- Weeks 41–44
26 Jul – 22 Aug 20274 · Signature project
- Read
- Fundamentals of Software Architecture (Richards, Ford); reading on the industry I pick
- Courses
- Optional: fast.ai Practical Deep Learning, part 1
- Build
- AI operations platform v1: agents read logs, traces, and alerts, and propose fixes
- Publish
- An article on the domain and the architecture
- Weeks 45–48
23 Aug – 19 Sep 20274 · Signature project
- Read
- The Staff Engineer's Path (Tanya Reilly); Software Architecture: The Hard Parts
- Courses
- Google Technical Writing One and Two; System Design Interview vol. 1 (Alex Xu)
- Build
- AI operations platform v2: human approval for risky actions, evals, a cost dashboard; an open-source pull request
- Publish
- A meetup talk proposal; the flagship write-up
- Weeks 49–52
20 Sep – 17 Oct 20274 · Job search
- Read
- System Design Interview vol. 2
- Courses
- NeetCode 150 (algorithm practice, about 30 minutes a day)
- Build
- A polished portfolio; two mock design interviews
- Publish
- The talk; applications to Staff and senior platform and AI roles
Courses from my subscriptions
One course per block. Each one replaces or deepens the course in the plan above, not both. Udemy for hands-on work and exam prep, Pluralsight's Skill IQ tests to skip what I already know, Udacity for one structured program with reviewed projects, and LinkedIn Learning for communication and leadership.
- Weeks 1–8
- Best pick
- Udemy: Ed Donner, AI Engineer Core Track: LLM Engineering, RAG, QLoRA, Agents
- Alternative
- Udacity: Agentic AI Nanodegree for weeks 1–16, as one structured program instead
- Weeks 9–12
- Best pick
- Udemy: Stephen Grider, Go: The Complete Developer's Guide (Golang)
- Alternative
- Pluralsight: Go skill path, after its Skill IQ test
- Weeks 13–16
- Best pick
- Udemy: Ed Donner, AI Engineer Agentic Track: The Complete Agent & MCP Course
- Alternative
- Covered by the Udacity Nanodegree, if I take it
- Weeks 17–20
- Best pick
- Udemy: Stephane Maarek, Apache Kafka Series: Learn Apache Kafka for Beginners
- Alternative
- Pluralsight: Kafka courses
- Weeks 21–24
- Best pick
- Udemy: Stephane Maarek, Ultimate AWS Certified Solutions Architect Associate
- Alternative
- Pluralsight: Kubernetes path, for the cluster work
- Weeks 25–28
- Best pick
- Udemy: Stephane Maarek, Solutions Architect Associate practice exams
- Alternative
- Udemy: OpenTelemetry Certified Associate practice tests (optional)
- Weeks 29–32
- Best pick
- LinkedIn Learning: Learning the OWASP Top 10, then the per-item OWASP Top 10 series
- Alternative
- Pluralsight: threat modeling courses
- Weeks 33–36
- Best pick
- Udemy: Jason Dion, CompTIA Security+ (SY0-701) complete course
- Alternative
- Pluralsight: Security+ path
- Weeks 37–40
- Best pick
- Udemy: Jason Dion, Security+ practice exams
- Alternative
- Udemy: KodeKloud (Mumshad Mannambeth), Certified Kubernetes Security Specialist (optional)
- Weeks 41–44
- Best pick
- LinkedIn Learning: Software Architecture Foundations
- Alternative
- Udemy: a deep learning course, only if I skip fast.ai
- Weeks 45–48
- Best pick
- LinkedIn Learning: technical writing and presenting to executives
- Alternative
- Udemy: Frank Kane, Mastering the System Design Interview
- Weeks 49–52
- Best pick
- LinkedIn Learning: salary negotiation and interviewing courses
- Alternative
- Udemy: an algorithms course, only if NeetCode is not enough
Course titles change between editions; search the title on the platform.
Every week
These run alongside the plan above.
- Katas, with AI off
Tuesdays. Rotating refactoring (Gilded Rose, Tennis), debugging (break a program, find the bug), and reviewing AI-written solutions. From week 17, one architectural kata a month.
- Writing
One design note per feature, and an article every two to three weeks: about twenty over the year.
- Career
Fridays. Two or three targeted applications or networking notes a week, and LinkedIn updated with each published piece.
- Monthly check
The last Sunday of each block: is the block's publish item out? If not, the next block shrinks rather than skipping it.
Resources and costs
Most courses are free. The books are cheapest through one O'Reilly Learning subscription (about $500 a year), which includes nearly all of them. Prices are approximate.