AI Engineering Receipts

Proof, not claims. Every artifact here is a clickable receipt from real systems I built and run: a 50-agent orchestration platform (MAX HQ), a live voice-AI SaaS (GetCallBounce), and an ML-augmented analytics platform (CitiPoints — public). The flagship repos are private (credentials, customer data, employer-adjacent IP), so this repo holds the genericized evidence: case studies, eval logs, architecture diagrams, CI runs, before/after numbers.

Interview prep (public, for any LLM)

File	Purpose	Use with
INTERVIEW-PLAYBOOK.md	The full FAANG AI Engineer interview playbook — methodology, 12 resume lines per-company reads, 60-question Q-bank slice, 30-day project plan, Q&A.	Humans + LLMs
INTERVIEW-METHODOLOGY.md	The first-principles methodology, standalone. 5 steps applied to every question.	Humans + LLMs
INTERVIEW-QBANK-LITE.json	100-question machine-readable Q-bank (1 per topic × difficulty, plus top FAANG-tagged). For LLM ingestion.	LLMs
INTERVIEW-HANDOFF.json	Self-contained brain — methodology + playbook + 100 questions + 12 resume lines + system prompt. One URL, any LLM.	LLMs

To make any LLM your FAANG interview mentor:

1. Open a chat with Claude, GPT, Gemini, or Grok.
2. Paste this URL: https://raw.githubusercontent.com/GopalGB/ai-engineering-receipts/main/INTERVIEW-HANDOFF.json
3. Say: "Read this. From now on, you are ARIA. Teach me."
4. The LLM now has the full methodology, the Q-bank, your resume context, and the system prompt.
5. Pick a question. It quizzes you. The loop continues until mastery or offer.

The live mentor (full 408-question bank, resume → Q mapping, Ask ARIA chat panel) is at /mentor in the career-os repo.

What's here

Folder	Receipt type
case-studies/	Architecture + hard-parts write-ups: MAX HQ, GetCallBounce, CitiPoints
evals/	Eval sets (evals.jsonl), LLM-judge scores, before/after averages
days/	Daily artifacts from the 32-day hardening sprint (tests, retry logic, tracing configs)
diagrams/	RAG service, multi-tenant agent orchestration, LLM gateway, agent loop

Already-public code (own work, not forks)

• citipoints — ML retail loyalty analytics, Next.js 15 + FastAPI
• rag-analytics-assistant — guardrailed LLM-first RAG over your own data
• codex-upgrade — portable power-kit for the Codex CLI (288 skills)
• officehub — self-hosted Notion-like workspace
• 6 merged upstream code PRs (linked per case study where relevant)

Rules this repo lives by

1. No secrets, ever — everything passes gitleaks + the pre-push gate before it lands.
2. No employer IP — employer work-product is never published; where a pattern matters, it's re-implemented clean-room on synthetic data (the CitiPoints pattern).
3. Forks are not featured — only code I wrote counts as a receipt.
4. Every claim carries a number — cost −X%, latency pY, eval avg before → after.