Skip to content

Apic Applied AI Architect — Interview Prep

A private, 12-module preparation system for the Apic Applied AI Architect role (Mumbai, Pre-Sales IC, India/APAC). Mirrors the structure of learngenai.sunmintz.com, adapted for interview-prep voice and Apic-bar grading.

What this course is

This is not a generic Solutions Architect course. It is a hyper-specific, role-calibrated prep system that:

  1. Executes the 12-module syllabus from the source meta-prompt (kept locally, not in repo)
  2. Layers in real Apic-specific calibration — current Claude lineup (Opus 4.7, Sonnet 4.6, Haiku 4.5), real Apic research references (Constitutional AI, Responsible Scaling Policy, interpretability), real Indian BFSI context (RBI, DPDPA 2023, hyperscaler India regions)
  3. Grades your drill answers honestly — Strong Hire / Hire / Lean No / Strong No, no softening
  4. Threads a single BFSI customer case study through every module so architecture skills compound, not fragment
  5. Sequences the Claude API artifact build into Module 04 — because zero hands-on Claude is the single largest credibility gap to close before applying

The course is private prep, not for publishing. Generic content will later be merged into a separate sun-course-solarch (AI Solutions Architect) course.

The 12 modules

Each module: Notes/ (concepts) + Drills/ (graded practice) + Case-Study/ (BFSI chapter) + Interview-QA-Bank.md (~26 questions). Module 04 adds CodeLabs/. Modules 05, 07, 08 add SystemDesigns/.

# Module What this module tests
01 Apic Calibration Whether you've internalized Apic's mission as operating mode, not vocabulary. The recruiter screen + initial values probe live here.
02 Personal Narrative & Role-Fit Whether your career arc tells a coherent story that lands you in this seat — and whether you own the IC lateral move without defensiveness.
03 Customer Discovery & Pre-Sales Whether you can run a discovery call with a regulated enterprise and translate ambiguous business asks into shippable architecture.
04 Claude Platform Architecture Whether you understand Claude's actual surface area — API, Claude for Work, Bedrock, Vertex, model selection, prompt caching, tool use. Includes the GitHub artifact build.
05 Solution Architecture Patterns Whether you can design enterprise-grade architectures for the six BFSI use cases — HLD, LLD, integration boundaries, failure modes.
06 Evaluation Architecture Whether you treat evals as the production go/no-go gate they are — not as an afterthought. Apic weighs this heavily.
07 RAG in Regulated Enterprises Whether you can architect retrieval that passes RBI scrutiny — chunking, embeddings, security boundaries, residency, hallucination control.
08 Agentic & Tool-Use Architecture Whether you can design agentic systems with safe tool schemas, MCP integration, and human-in-the-loop approval — and explain why automation has limits.
09 Security, Privacy, Governance Whether a CISO would trust you. Prompt injection, data leakage, audit, role-based access, regulated-workload posture.
10 Cost, Latency, Reliability Whether you treat production AI as a system, not a demo — observability, cost-per-BU, latency budgets, fallback strategy.
11 Storytelling, Workshops, Demos Whether you can speak the same architecture in five registers — CEO, CIO, CTO, CISO, hands-on engineer — without losing technical credibility or business clarity.
12 Full Loop Simulation Whether the prep adds up. Mock recruiter → HM → onsite Loop 1 (technical) → onsite Loop 2 (values + project deep dive) — all graded.

How to use this course

Sequential. Modules build on each other. Module 04's CodeLabs require Module 03's discovery framing to make sense. Module 11's executive storytelling needs Modules 05–10's architecture depth as raw material.

Drill-driven. Every module ends with practice drills you actually perform — written, spoken aloud, or whiteboarded. Your answers get graded against a Strong Hire / Hire / Lean No / Strong No rubric. Grades + upgraded versions are logged in Drill Tracker.

Case-study threaded. A single Indian BFSI customer (six use cases, real constraints) appears in every module's Case-Study/ chapter. By Module 12, you have a complete enterprise architecture portfolio for one customer — far more memorable in an interview than five disconnected toy examples.

Apic-calibrated. Every "good answer" example references actual Apic research, the actual JD language, the actual Claude model lineup. Generic AI talk gets a Lean No, by design.

Application trigger checklist

You do not apply until all four are true:

  • [ ] Apic-tailored resume .docx written and reviewed (Module 02 deliverable)
  • [ ] Claude API artifact published on GitHub — Python SDK + tool use + structured output + prompt caching + custom eval harness (Module 04 deliverable)
  • [ ] "Tell me about yourself" + "Why Apic" both at Strong Hire (Module 01 drills)
  • [ ] Clean lateral-move answer — non-defensive, principal-grade narrative (Module 02 deliverable)

The fit analysis flagged "apply this week" — but speed matters less than these four boxes. The artifact alone is 4–6 hours of focused work.

Sources

  • Meta-prompt (the source of truth): kept locally, not committed
  • Job description: kept locally, not committed
  • Fit analysis: kept locally, not committed
  • Current resume: kept locally, not committed
  • Public interview intel: techprep.app + jobright.ai (covered SWE/RS/MLE; loop structure transfers, technical-round content does not)
  • Approved plan: vibes/plan.md (in repo, outside the rendered docs site)
  • Repo context for future agents: vibes/CLAUDE.md
  • Session handoff: vibes/handoff.md