# Build Production-Ready LLM Systems

Production AI systems require reliable retrieval, orchestration, safety, monitoring, and cost controls. This skill provides structured engineering guidance across the complete LLM application lifecycle.

## Install

```bash
npx skillstore add sickn33/ai-engineer
```

## Metadata

- Status: approved
- Slug: sickn33-ai-engineer
- Skillstore revision: r2
- Version status: missing
- Tree hash: 84c44168a3d7da12b53a80e9697622eae0ca383765639e49210dd3251a0a50d1
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/ai-engineer
- Ref: ebdfe608f5de2b66ff37ab4af12af8ac4f5e8006
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Quality score: 78
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/sickn33-ai-engineer
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-ai-engineer/manifest

## Capabilities

- Designs multi-stage RAG pipelines with chunking, hybrid retrieval, reranking, and context filtering.
- Recommends models, embedding services, vector databases, and agent frameworks based on cost, latency, reliability, and scale.
- Plans production API patterns with streaming, caching, rate limits, fallbacks, and structured outputs.
- Designs agent workflows with tools, memory, state management, escalation paths, and evaluation metrics.
- Defines safeguards for prompt injection, sensitive data, content moderation, bias, and policy compliance.
- Creates testing, monitoring, cost management, and staged rollout strategies for AI services.

## Use Cases

- Add an LLM Feature: Design a reliable model integration with typed outputs, fallback behavior, testing, and cost limits.
- Build Enterprise RAG: Plan ingestion, chunking, hybrid retrieval, reranking, evaluation, access control, and production monitoring.
- Define an Agent Architecture: Turn a product workflow into tools, state transitions, guardrails, human escalation, and measurable success criteria.

## Prompt Templates

### Plan a Basic LLM Feature

```
Design an LLM feature for [use case]. State assumptions, model choice, request flow, safety controls, tests, and a small rollout plan.
```

### Design a RAG Pipeline

```
Design a RAG system for [documents and users]. Cover ingestion, chunking, retrieval, reranking, citations, evaluation, permissions, latency, and cost.
```

### Architect an AI Agent

```
Design an agent for [workflow]. Define tools, permissions, state, memory, failure handling, human approval points, observability, and adversarial tests.
```

### Review a Production AI Platform

```
Review this architecture: [details]. Identify reliability, security, quality, latency, and cost risks. Prioritize fixes and define measurable release gates.
```

## Limitations

- Generated designs and code require environment-specific testing, security review, and operational validation.
- The skill does not provide credentials, datasets, infrastructure access, or deployment permissions.
- Model and framework recommendations can become outdated as providers change features, pricing, and availability.
- Reliable results require clear use cases, data sources, success metrics, constraints, and safety boundaries.

## Best Practices

- Define quality, latency, cost, safety, and reliability targets before selecting models or frameworks.
- Evaluate retrieval and generation separately with representative data, adversarial inputs, and traceable failure categories.
- Use least-privilege tools, redact sensitive data, require approval for high-impact actions, and monitor production behavior.

## Anti Patterns

- Do not choose an agent when a deterministic workflow or simple model call can solve the task.
- Do not send sensitive data to external models without authorization, minimization, redaction, and retention controls.
- Do not launch without evaluation baselines, observability, fallback behavior, cost limits, and a rollback plan.

## Security Audit

- Audited at: 2026-07-23T20:41:11.192\+00:00
- Summary: All four static findings are false positives caused by safety and information-retrieval terminology in SKILL.md. The file contains guidance only and does not direct system reconnaissance, jailbreak behavior, code execution, or data exfiltration. No net-new semantic security risks were found.

## Stats

- Views: 92
- Downloads: 8
- Favorites: 0
- Popularity score: 0