Build it · break it · explain it · ship it · apply. A battle-tested, day-by-day journey to Forward Deployed Engineer, Applied AI, and GenAI roles. Master enterprise architectures, write hand-crafted core modules, inject deliberate faults, and ship production-grade systems.
Four foundational principles engineered to turn software engineers into autonomous FDEs.
Each concept block connects theory directly to production code. Core Python is written by hand in weeks 1–3. From week 4 onwards, AI-generated code only counts once you can explain, test, and debug every line under three consecutive “why?”s.
Every build week includes an adversarial Break-It lab: state your hypothesis → inject the fault → observe the failure modes → diagnose root causes → fix and lock in with automated regression tests. Every lab feeds your technical story bank.
Five rigorous checks scored 0–2: Explain, Compare, Diagram, Apply, Defend. A pass requires ≥ 8/10 with Apply = 2. Four unskippable hard gates: PII + prompt injection, tool permissions, IAM least privilege, and tenant isolation.
A strategic, tiered job search: safety-net roles anytime; Tier C applications from week 3; Tier B from week 5; and Tier A (AI labs, Palantir, Databricks, Google Cloud) from week 7 once SWIFTCLAIM v2 ships to production.
Consistent cadence designed for deep work without burnout.
| Learning Block | Mon–Fri (3 hrs) | Saturday (5 hrs) | Sunday (2 hrs) |
|---|---|---|---|
| Recall (closed book) | 10 min | 15 min | — |
| Concept Deep Dive | 55 min | 60 min | — |
| Build — SWIFTCLAIM | 70 min | 135 min | — |
| Coding Drill (timed, no AI) | 25 min | — | — |
| FDE Rep / Case Study | — | 45 min | — |
| Job Search & Outreach | 20 min | 45 min | — |
| Gate Defense & Gap Audit | — | — | 120 min |
An enterprise AI-assisted claims-triage copilot for insurance carriers, built on synthetic datasets in one public repository.
A production-shaped API service that ingests incident files, performs hybrid keyword and vector retrieval via pgvector, and generates triage assessments with strict citation grounding and abstention logic.
A Go-based Model Context Protocol (MCP) server enabling a Bedrock-orchestrated agent with human-in-the-loop approvals, full Terraform AWS infra, multi-tenant isolation, automated rollbacks, and OpenTelemetry tracing.
Published days open directly. Locked days show the planned syllabus and unlock as you advance.
Know which claims you can defend, explain LLM mechanics without buzzwords, and get SWIFTCLAIM returning validated JSON from Claude.
Confirm certified knowledge fast, then make triage production-shaped: retries, timeouts, cost, tools and a first accuracy number.
Build and diagnose a RAG pipeline end to end, and answer “why this, not that?” for every component.
Prove quality with evidence, defend against injection and leakage, and ship v1.
An agent whose permissions are enforced in code, not in the prompt — the flagship FDE story.
Deploy SWIFTCLAIM securely on AWS and defend every network, identity and data decision.
Must-ship v2 work first, stretch work in the agreed cut order. Ship v2 and defend the whole architecture under pressure.
Walk from customer problem to production, and defend every decision, story and number under interview conditions.
Close the gap list, rehearse under pressure, and enter January with a live pipeline.
Engineering principles that protect your integrity and interview defense.
The story bank holds only work you personally delivered, at the exact scope you delivered it. SWIFTCLAIM is always labeled “personal project, synthetic data”.
Formulate and document your own hypothesis before consulting an LLM. Treat AI suggestions as unproven until an automated test confirms them.
Coding drills are strictly closed-book and no-AI unless explicitly marked. Explain time and space complexity aloud before executing tests.
Sunday gates are non-negotiable. Missed checks are appended to the gap list and re-tested the following Sunday until mastered.
One model serving provider until week 6, one vector store, and strictly no frontend beyond OpenAPI/CLI to maintain relentless focus on systems engineering.
90 minutes maximum per Break-It lab. Stuck for 60 minutes on a dependency or tool? Move one rung down its prescribed fallback ladder.