project-development

LLM Project Development

Design and build LLM-powered projects from ideation through deployment. This skill covers the meta-level decisions that shape a project before any code is written: whether LLMs are appropriate for the task, how to structure the pipeline, what output formats to use, how to handle batch processing, and how to plan for cost and scale.

When to Activate

Activate this skill when:

• Starting a new LLM-powered project from scratch
• Analyzing whether a task is a good fit for language models
• Designing pipeline architecture (single-pass, chain, DAG, agent loop)
• Choosing structured output schemas and validation strategies
• Planning batch processing for throughput optimization
• Estimating costs and planning resource allocation

Do not activate this skill for adjacent work owned by other skills:

• Designing individual tool schemas and descriptions: tool-design.
• Choosing multi-agent topology (supervisor vs. swarm): multi-agent-patterns.
• Building evaluation pipelines and metrics: evaluation.
• Designing agent operating loops: harness-engineering.

Core Concepts

Start every project with task-model fit analysis. Not all tasks benefit from LLMs — some are better served by traditional software, rules engines, or simpler ML models. LLMs excel at tasks requiring language understanding, generation, reasoning across unstructured data, and flexible decision-making. They struggle with tasks requiring exact arithmetic, deterministic logic, real-time performance, or guaranteed correctness.

Design pipelines around the complexity of the task, not the sophistication of the technology:

Pipeline Type	When to Use	Complexity
Single-pass	Task has one clear input → output	Lowest
Chain	Task decomposes into sequential steps	Low
DAG	Steps have conditional branches or parallel paths	Medium
Agent loop	Task requires exploration, tool use, or self-correction	Highest

Choose the simplest pipeline that solves the problem. Over-engineering adds cost, latency, and debugging difficulty without proportional quality gains.