CCA Exam Study Guide — Claude Certified Architect Foundations

1. Exam Overview

The Claude Certified Architect – Foundations certification validates that practitioners can make informed decisions about tradeoffs when implementing real-world solutions with Claude. Questions are grounded in realistic scenarios drawn from actual customer use cases.

1.1 Format & Scoring

1.2 Domain Weights

pie title CCA Exam Domain Distribution
    "D1: Agentic Architecture (27%)" : 27
    "D2: Tool Design & MCP (18%)" : 18
    "D3: Claude Code Config (20%)" : 20
    "D4: Prompt Engineering (20%)" : 20
    "D5: Context & Reliability (15%)" : 15
                        

2. Exam Scenarios

The exam presents 4 of 6 scenarios selected at random. Each scenario provides a realistic production context that frames a set of questions.

3. Domain 1: Agentic Architecture & Orchestration (27%)

Task 1.1: Design and implement agentic loops for autonomous task execution

Knowledge of:

• The agentic loop lifecycle: sending requests, inspecting stop_reason ("tool_use" vs "end_turn"), executing tools, and returning results Part 3 §1.1
• How tool results are appended to conversation history for the next iteration Part 3 §1.3
• Model-driven decision-making vs pre-configured decision trees or tool sequences

Skills in:

• Implementing control flow that continues on "tool_use" and terminates on "end_turn" Part 3 §2
• Adding tool results to conversation context between iterations Part 3 §1.3
• Avoiding anti-patterns: parsing NL signals, arbitrary iteration caps, checking text content for completion Part 3 §3

Task 1.2: Orchestrate multi-agent systems with coordinator-subagent patterns

Knowledge of:

• Hub-and-spoke architecture where a coordinator manages all inter-subagent communication Part 4 §1.1
• Subagents operate with isolated context — they do NOT inherit the coordinator's history Part 4 §1.2
• Coordinator's role in task decomposition, delegation, result aggregation, and subagent selection Part 4 §1.1
• Risks of overly narrow task decomposition leading to incomplete coverage Part 4 §1.3

Skills in:

• Designing coordinators that dynamically select subagents based on query requirements Part 4 §1.1
• Partitioning research scope to minimize duplication Part 4 §1.2
• Implementing iterative refinement loops (evaluate, re-delegate, re-synthesize) Part 4 §1.3
• Routing all communication through the coordinator for observability Part 4 §1.1

Task 1.3: Configure subagent invocation, context passing, and spawning

Knowledge of:

• The Task tool as the mechanism for spawning subagents; allowedTools must include "Task" Part 4 §2.1
• Subagent context must be explicitly provided in the prompt — no automatic inheritance Part 4 §2.5
• AgentDefinition configuration: descriptions, system prompts, tool restrictions Part 4 §2.1
• fork_session for creating independent branches from a shared analysis baseline Part 5 §2.3

Skills in:

• Including complete findings from prior agents directly in subagent prompts Part 4 §2.5
• Using structured data formats to separate content from metadata (source URLs, page numbers) Part 4 §2.5
• Spawning parallel subagents by emitting multiple Task tool calls in a single coordinator response Part 4 §2.6
• Designing coordinator prompts that specify goals and quality criteria rather than step-by-step procedures Part 4 §2.5

Task 1.4: Implement multi-step workflows with enforcement and handoff patterns

Knowledge of:

• Programmatic enforcement (hooks, prerequisite gates) vs prompt-based guidance for workflow ordering Part 4 §3.1
• Prompt instructions alone have a non-zero failure rate for deterministic compliance requirements Part 4 §3.1
• Structured handoff protocols for escalation: customer details, root cause, recommended actions Part 4 §3.3

Skills in:

• Implementing programmatic prerequisites that block downstream calls until prerequisites complete Part 4 §3.2
• Decomposing multi-concern requests into distinct items, investigating in parallel Part 4 §3.2
• Compiling structured handoff summaries for human escalation Part 4 §3.3 Part 18 §3.1

Task 1.5: Apply Agent SDK hooks for tool call interception and data normalization

Knowledge of:

• PostToolUse hooks that intercept tool results for transformation before the model processes them Part 5 §1.2
• Pre-tool hooks that intercept outgoing calls to enforce compliance (e.g., blocking refunds above a threshold) Part 5 §1.1
• Hooks for deterministic guarantees vs prompt instructions for probabilistic compliance Part 5 §1.4

Skills in:

• Implementing PostToolUse hooks to normalize heterogeneous data formats Part 5 §1.2
• Implementing tool call interception that blocks policy-violating actions and redirects to alternative workflows Part 5 §1.1
• Choosing hooks over prompts when business rules require guaranteed compliance Part 5 §1.4

Task 1.6: Design task decomposition strategies for complex workflows

Knowledge of:

• When to use fixed sequential pipelines (prompt chaining) vs dynamic adaptive decomposition Part 3 §4.3
• Prompt chaining patterns that break reviews into sequential steps Part 3 §4.1
• Adaptive investigation plans that generate subtasks based on intermediate discoveries Part 3 §4.2

Skills in:

• Selecting appropriate decomposition patterns: prompt chaining for predictable reviews, dynamic decomposition for open-ended investigation Part 3 §4.3
• Splitting large code reviews into per-file local passes plus a cross-file integration pass Part 3 §4.1
• Decomposing open-ended tasks by first mapping structure, then creating prioritized adaptive plans Part 3 §4.2

Task 1.7: Manage session state, resumption, and forking

Knowledge of:

• Named session resumption using --resume <session-name> Part 5 §2.2
• fork_session for creating independent branches from a shared baseline Part 5 §2.3
• Informing the agent about changes to previously analyzed files when resuming Part 5 §2.2
• Why starting fresh with a structured summary is more reliable than resuming with stale tool results Part 5 §2.4

Skills in:

• Using --resume with session names to continue named investigation sessions Part 5 §2.2
• Using fork_session for parallel exploration branches Part 5 §2.3
• Choosing between session resumption and fresh starts with injected summaries Part 5 §2.4

4. Domain 2: Tool Design & MCP Integration (18%)

Task 2.1: Design effective tool interfaces with clear descriptions and boundaries

Knowledge of:

• Tool descriptions as the primary mechanism LLMs use for tool selection; minimal descriptions lead to unreliable selection Part 6 §1.1
• Including input formats, example queries, edge cases, and boundary explanations Part 6 §1.1
• How ambiguous or overlapping descriptions cause misrouting Part 6 §3.1
• System prompt wording impact on tool selection: keyword-sensitive instructions create unintended associations Part 6 §4

Skills in:

• Writing descriptions that clearly differentiate each tool's purpose, expected inputs, outputs, and when to use it Part 6 §1.1
• Renaming tools and updating descriptions to eliminate functional overlap Part 6 §3.1
• Splitting generic tools into purpose-specific tools with defined contracts Part 6 §3.1

Task 2.2: Implement structured error responses for MCP tools

Knowledge of:

• The MCP isError flag for communicating tool failures Part 6 §2.1
• Distinguishing transient, validation, business, and permission errors Part 6 §2.2
• Why uniform errors ("Operation failed") prevent appropriate recovery Part 6 §2.2
• Retryable vs non-retryable errors and structured metadata to prevent wasted retries Part 6 §5.4

Skills in:

• Returning structured error metadata: errorCategory, isRetryable, human-readable descriptions Part 6 §2.2
• Including retriable: false flags for business rule violations Part 6 §5.4
• Implementing local error recovery within subagents for transient failures Part 6 §2.2
• Distinguishing access failures from valid empty results Part 6 §2.1

Task 2.3: Distribute tools appropriately across agents and configure tool choice

Knowledge of:

• Giving an agent too many tools (e.g., 18 instead of 4–5) degrades tool selection reliability Part 22 §4
• Agents with tools outside their specialization tend to misuse them Part 22 §4.2
• Scoped tool access: giving agents only the tools needed for their role Part 22 §4
• tool_choice options: "auto", "any", and forced selection ({"type": "tool", "name": "..."}) Part 17 §2.2

Skills in:

• Restricting each subagent's tool set to prevent cross-specialization misuse Part 22 §4.2
• Replacing generic tools with constrained alternatives Part 22 §4
• Using tool_choice forced selection to ensure a specific tool is called first Part 17 §2.2
• Setting tool_choice: "any" to guarantee tool calls over conversational text Part 17 §2.2

Task 2.4: Integrate MCP servers into Claude Code and agent workflows

Knowledge of:

• MCP server scoping: project-level (.mcp.json) vs user-level (~/.claude.json) Part 7 §5.3
• Environment variable expansion (${GITHUB_TOKEN}) for credential management Part 7 §5.3
• Tools from all configured MCP servers are discovered at connection time Part 7 §6.1
• MCP resources for exposing content catalogs to reduce exploratory tool calls Part 7 §6.1

Skills in:

• Configuring shared MCP servers in project-scoped .mcp.json Part 7 §5.3
• Configuring personal servers in user-scoped ~/.claude.json Part 7 §5.3
• Enhancing MCP tool descriptions to prevent the agent from preferring built-in tools Part 7 §6.1
• Choosing existing community MCP servers over custom implementations for standard integrations Part 7 §6.1

Task 2.5: Select and apply built-in tools (Read, Write, Edit, Bash, Grep, Glob) effectively

Knowledge of:

• Grep for content search (file contents for patterns); Glob for file path pattern matching Part 22 §1.2
• Read/Write for full file operations; Edit for targeted modifications using unique text matching Part 22 §1.1
• When Edit fails due to non-unique matches, use Read + Write as fallback Part 22 §1.1

Skills in:

• Selecting Grep for code content search, Glob for finding files by pattern Part 22 §1.2
• Building codebase understanding incrementally: Grep → Read → follow imports → trace flows Part 22 §1.2
• Tracing function usage across wrapper modules by identifying exported names, then searching each Part 22 §1.3

5. Domain 3: Claude Code Configuration & Workflows (20%)

Task 3.1: Configure CLAUDE.md files with appropriate hierarchy, scoping, and modular organization

Knowledge of:

• Hierarchy: user-level (~/.claude/CLAUDE.md), project-level (.claude/CLAUDE.md or root), directory-level Part 8 §1.1
• User-level settings apply only to that user — not shared via version control Part 8 §1.1
• @import syntax for referencing external files Part 8 §5.1
• .claude/rules/ directory for topic-specific rule files Part 8 §5.1

Skills in:

• Diagnosing configuration hierarchy issues (e.g., new team member not receiving instructions) Part 8 §1.1
• Using @import to selectively include relevant standards files per package Part 8 §5.1
• Splitting large CLAUDE.md files into focused files in .claude/rules/ Part 8 §5.1
• Using /memory to verify which memory files are loaded Part 8 §5.2

Task 3.2: Create and configure custom slash commands and skills

Knowledge of:

• Project-scoped commands in .claude/commands/ (shared) vs user-scoped in ~/.claude/commands/ (personal) Part 8 §5.3
• Skills in .claude/skills/ with SKILL.md supporting context: fork, allowed-tools, argument-hint Part 8 §3.2
• context: fork for running skills in isolated sub-agent context Part 8 §3.2

Skills in:

• Creating project-scoped slash commands for team-wide availability Part 8 §5.3
• Using context: fork to isolate verbose skill output Part 8 §3.2
• Configuring allowed-tools in skill frontmatter to restrict tool access Part 8 §3.2
• Using argument-hint to prompt for required parameters Part 8 §3.2

Task 3.3: Apply path-specific rules for conditional convention loading

Knowledge of:

• .claude/rules/ files with YAML frontmatter paths fields containing glob patterns Part 8 §5.1
• Path-scoped rules load only when editing matching files, reducing irrelevant context Part 8 §5.1
• Glob-pattern rules vs directory-level CLAUDE.md for conventions spanning multiple directories Part 8 §5.1

Skills in:

• Creating files with path scoping (e.g., paths: ["terraform/**/*"]) Part 8 §5.1
• Using glob patterns for file-type conventions regardless of location (e.g., **/*.test.tsx) Part 8 §5.1
• Choosing path-specific rules over subdirectory CLAUDE.md for cross-directory conventions Part 8 §5.1

Task 3.4: Determine when to use plan mode vs direct execution

Knowledge of:

• Plan mode for complex tasks: large-scale changes, multiple valid approaches, architectural decisions Part 9 §1
• Direct execution for simple, well-scoped changes Part 9 §1.1
• Plan mode enables safe exploration before committing to changes Part 9 §1
• Explore subagent for isolating verbose discovery output Part 9 §1.1

Skills in:

• Selecting plan mode for architectural implications (microservice restructuring, 45+ file migrations) Part 9 §1
• Selecting direct execution for well-understood changes with clear scope Part 9 §1.1
• Using the Explore subagent for verbose discovery phases Part 9 §1.1
• Combining plan mode for investigation with direct execution for implementation Part 9 §1

Task 3.5: Apply iterative refinement techniques for progressive improvement

Knowledge of:

• Concrete input/output examples as the most effective way to communicate transformations Part 10 §2
• Test-driven iteration: write tests first, iterate by sharing test failures Part 9 §2.1
• The interview pattern: having Claude ask questions to surface considerations Part 10 §2
• When to provide all issues at once (interacting) vs sequentially (independent) Part 9 §2.1

Skills in:

• Providing 2–3 concrete input/output examples to clarify transformation requirements Part 10 §2
• Writing test suites first, then iterating by sharing test failures Part 9 §2.1
• Using the interview pattern for unfamiliar domains Part 10 §2
• Addressing interacting issues in one message vs sequential iteration for independent issues Part 9 §2.1

Task 3.6: Integrate Claude Code into CI/CD pipelines

Knowledge of:

• The -p (or --print) flag for non-interactive mode in automated pipelines Part 9 §4.1
• --output-format json and --json-schema for structured CI output Part 9 §4.2
• CLAUDE.md provides project context to CI-invoked Claude Code Part 9 §4.3
• Session context isolation: same session that generated code is less effective at self-review Part 9 §4.3

Skills in:

• Running Claude Code in CI with -p flag Part 9 §4.1
• Using --output-format json with --json-schema for structured PR comments Part 9 §4.2
• Including prior review findings in context to avoid duplicate comments Part 9 §4.3
• Documenting testing standards in CLAUDE.md for better test generation Part 9 §4.3

6. Domain 4: Prompt Engineering & Structured Output (20%)

Task 4.1: Design prompts with explicit criteria to improve precision and reduce false positives

Knowledge of:

• Explicit criteria over vague instructions (e.g., "flag comments only when claimed behavior contradicts actual code" vs "check comments are accurate") Part 10 §1
• General instructions like "be conservative" fail to improve precision compared to specific categorical criteria Part 10 §1
• High false positive rates undermine developer trust in accurate categories Part 10 §1

Skills in:

• Writing specific review criteria defining which issues to report vs skip Part 10 §1
• Temporarily disabling high false-positive categories to restore trust Part 10 §1
• Defining explicit severity criteria with concrete code examples Part 10 §1

Task 4.2: Apply few-shot prompting to improve output consistency and quality

Knowledge of:

• Few-shot examples as the most effective technique for consistently formatted output Part 10 §2
• Few-shot examples for demonstrating ambiguous-case handling Part 10 §2
• How few-shot examples enable generalization to novel patterns Part 10 §2
• Effectiveness for reducing hallucination in extraction tasks Part 10 §2

Skills in:

• Creating 2–4 targeted few-shot examples for ambiguous scenarios with reasoning Part 10 §2
• Including examples that demonstrate the desired output format Part 10 §2
• Providing examples distinguishing acceptable patterns from genuine issues Part 10 §2
• Adding examples showing correct extraction from varied document formats Part 10 §2

Task 4.3: Enforce structured output using tool use and JSON schemas

Knowledge of:

• tool_use with JSON schemas as the most reliable approach for guaranteed schema-compliant output Part 11 §1
• tool_choice modes: "auto" (may return text), "any" (must call a tool), forced (must call specific tool) Part 11 §5.1
• Strict JSON schemas eliminate syntax errors but NOT semantic errors (values in wrong fields) Part 11 §1
• Schema design: required vs optional fields, enum + "other" + detail string patterns Part 11 §1

Skills in:

• Defining extraction tools with JSON schemas and extracting from tool_use response Part 11 §1
• Setting tool_choice: "any" for guaranteed structured output with multiple schemas Part 11 §5.1
• Forcing a specific tool with tool_choice: {"type": "tool", "name": "..."} Part 17 §2.2
• Designing optional/nullable fields to prevent model fabrication Part 11 §1
• Adding enum values like "unclear" for ambiguous cases Part 11 §1

Task 4.4: Implement validation, retry, and feedback loops for extraction quality

Knowledge of:

• Retry-with-error-feedback: appending specific validation errors to the prompt on retry Part 11 §2
• Retries are ineffective when information is absent from the source document Part 11 §4
• Feedback loop design: tracking detected_pattern field for dismissal analysis Part 11 §4
• Semantic validation errors (values don't sum) vs schema syntax errors (eliminated by tool use) Part 11 §2

Skills in:

• Implementing follow-up requests with original document, failed extraction, and specific validation errors Part 11 §2
• Identifying when retries will be ineffective vs when they will succeed Part 11 §4
• Adding detected_pattern fields for false positive analysis Part 13 §1
• Designing self-correction: extracting "calculated_total" alongside "stated_total" to flag discrepancies Part 11 §2

Task 4.5: Design efficient batch processing strategies

Knowledge of:

• Message Batches API: 50% cost savings, up to 24-hour processing, no latency SLA Part 11 §3
• Appropriate for non-blocking workloads (overnight reports); inappropriate for blocking workflows (pre-merge checks) Part 11 §3
• Batch API does not support multi-turn tool calling within a single request Part 11 §3
• custom_id fields for correlating request/response pairs Part 11 §3

Skills in:

• Matching API approach to workflow latency requirements: synchronous for blocking, batch for overnight Part 11 §3
• Calculating batch submission frequency based on SLA constraints Part 11 §3
• Handling batch failures: resubmitting only failed documents by custom_id Part 11 §3
• Using prompt refinement on a sample set before batch-processing large volumes Part 11 §3

Task 4.6: Design multi-instance and multi-pass review architectures

Knowledge of:

• Self-review limitations: model retains reasoning context, making it less likely to question its own decisions Part 11 §6.1
• Independent review instances are more effective than self-review or extended thinking Part 11 §6.1
• Multi-pass review: per-file local passes plus cross-file integration passes Part 11 §6.2

Skills in:

• Using a second independent Claude instance to review generated code without the generator's reasoning context Part 11 §6.1
• Splitting large multi-file reviews into focused per-file passes plus integration passes Part 11 §6.2
• Running verification passes with self-reported confidence for calibrated routing Part 11 §6.2

7. Domain 5: Context Management & Reliability (15%)

Task 5.1: Manage conversation context to preserve critical information across long interactions

Knowledge of:

• Progressive summarization risks: condensing numerical values, dates, expectations into vague summaries Part 12 §2
• "Lost in the middle" effect: models reliably process beginnings and ends but may omit middle sections Part 12 §6.2
• Tool results accumulate and consume tokens disproportionately to their relevance Part 12 §1
• Complete conversation history must be passed in subsequent API requests Part 12 §1

Skills in:

• Extracting transactional facts into a persistent "case facts" block Part 12 §1
• Trimming verbose tool outputs to only relevant fields Part 12 §1
• Placing key findings summaries at the beginning of aggregated inputs Part 12 §6.2
• Modifying upstream agents to return structured data instead of verbose content Part 12 §6.3

Task 5.2: Design effective escalation and ambiguity resolution patterns

Knowledge of:

• Appropriate escalation triggers: customer requests for a human, policy exceptions/gaps, inability to progress Part 13 §2
• Escalating immediately when customer explicitly demands it vs offering to resolve straightforward issues Part 18 §4.1
• Sentiment-based escalation and self-reported confidence scores are unreliable proxies Part 13 §2
• Multiple customer matches require clarification, not heuristic selection Part 18 §4.1

Skills in:

• Adding explicit escalation criteria with few-shot examples Part 13 §2
• Honoring explicit customer requests for human agents immediately Part 18 §4.1
• Escalating when policy is ambiguous or silent on the customer's request Part 13 §2
• Asking for additional identifiers when tool results return multiple matches Part 18 §4.1

Task 5.3: Implement error propagation strategies across multi-agent systems

Knowledge of:

• Structured error context (failure type, attempted query, partial results, alternatives) enables intelligent coordinator recovery Part 13 §1
• Distinguishing access failures (timeouts needing retry) from valid empty results Part 6 §2.2
• Generic error statuses hide valuable context from the coordinator Part 13 §1
• Silently suppressing errors or terminating entire workflows on single failures are both anti-patterns Part 13 §1

Skills in:

• Returning structured error context including failure type, what was attempted, partial results Part 13 §1
• Having subagents implement local recovery for transient failures Part 13 §1
• Structuring synthesis output with coverage annotations indicating gaps Part 6 §2.2

Task 5.4: Manage context effectively in large codebase exploration

Knowledge of:

• Context degradation in extended sessions: inconsistent answers, "typical patterns" references Part 12 §6.1
• Scratchpad files for persisting key findings across context boundaries Part 12 §6.3
• Subagent delegation for isolating verbose exploration output Part 12 §6.1
• Structured state persistence for crash recovery: agent exports state to manifest files Part 23 §3

Skills in:

• Spawning subagents for specific investigation questions Part 12 §6.1
• Maintaining scratchpad files recording key findings Part 12 §6.3
• Summarizing findings before spawning subagents for the next phase Part 12 §6.1
• Using /compact to reduce context during extended sessions Part 12 §6.1

Task 5.5: Design human review workflows and confidence calibration

Knowledge of:

• Aggregate accuracy metrics (97% overall) may mask poor performance on specific document types Part 13 §7
• Stratified random sampling for measuring error rates in high-confidence extractions Part 13 §7
• Field-level confidence scores calibrated using labeled validation sets Part 13 §5.1
• Validate accuracy by document type and field before automating high-confidence extractions Part 13 §7

Skills in:

• Implementing stratified random sampling of high-confidence extractions Part 13 §7
• Analyzing accuracy by document type and field Part 13 §5.1
• Having models output field-level confidence scores Part 13 §5.1
• Routing low-confidence extractions to human review Part 13 §7

Task 5.6: Preserve information provenance and handle uncertainty in multi-source synthesis

Knowledge of:

• Source attribution lost during summarization without preserving claim-source mappings Part 20 §2
• Structured claim-source mappings that synthesis agents must preserve and merge Part 20 §2
• Handling conflicting statistics: annotating conflicts with source attribution, not selecting one arbitrarily Part 20 §2.2
• Temporal data: requiring publication/collection dates to prevent misinterpreted contradictions Part 4 §2.5

Skills in:

• Requiring subagents to output structured claim-source mappings Part 20 §2
• Structuring reports distinguishing well-established from contested findings Part 20 §2.2
• Requiring publication/collection dates in structured outputs Part 4 §2.5
• Rendering different content types appropriately (financial data as tables, news as prose) Part 20 §2

8. Sample Questions

The following 12 sample questions illustrate the format and difficulty level of the CCA exam. Each includes the correct answer and explanation.

Scenario: Customer Support Resolution Agent

Scenario: Code Generation with Claude Code

Scenario: Multi-Agent Research System

Scenario: Claude Code for Continuous Integration

9. Preparation Exercises

10. Appendix

Technologies & Concepts on the Exam

• Claude Agent SDK — Agent definitions, agentic loops, stop_reason handling, hooks (PostToolUse, tool call interception), subagent spawning via Task tool, allowedTools
• Model Context Protocol (MCP) — MCP servers, tools, resources, isError flag, tool descriptions, .mcp.json, environment variable expansion
• Claude Code — CLAUDE.md hierarchy, .claude/rules/ with YAML frontmatter, .claude/commands/, .claude/skills/ with context: fork/allowed-tools/argument-hint, plan mode, /memory, /compact, --resume, fork_session, Explore subagent
• Claude Code CLI — -p / --print flag, --output-format json, --json-schema
• Claude API — tool_use with JSON schemas, tool_choice ("auto", "any", forced), stop_reason values, max_tokens, system prompts
• Message Batches API — 50% cost savings, 24-hour window, custom_id, no multi-turn tool calling
• JSON Schema — Required/optional fields, enum types, nullable, "other" + detail patterns, strict mode
• Pydantic — Schema validation, semantic validation, retry loops
• Built-in tools — Read, Write, Edit, Bash, Grep, Glob
• Few-shot prompting — Targeted examples, format demonstration, generalization
• Prompt chaining — Sequential decomposition into focused passes
• Context window — Token budgets, progressive summarization, lost-in-the-middle, scratchpad files
• Session management — Resumption, fork_session, named sessions, isolation
• Confidence scoring — Field-level confidence, calibration with validation sets, stratified sampling

Out of Scope (NOT on the exam)

• Fine-tuning or training custom models
• API authentication, billing, or account management
• Deploying/hosting MCP servers (infrastructure, networking, containers)
• Claude's internal architecture, training, or weights
• Constitutional AI, RLHF, or safety training methodologies
• Embedding models or vector database implementation
• Computer use (browser automation, desktop interaction)
• Vision/image analysis capabilities
• Streaming API implementation or server-sent events
• Rate limiting, quotas, or API pricing calculations
• OAuth, API key rotation, or authentication protocols
• Cloud provider configurations (AWS, GCP, Azure)
• Performance benchmarking or model comparison metrics
• Prompt caching implementation details
• Token counting algorithms or tokenization specifics