Glossary Q - T

Technique combining 4-bit quantization with LoRA to enable fine-tuning large language models on consumer hardware, democratizing LLM adaptation.

Technique that reduces the numerical precision of AI model weights to decrease size and accelerate inference, with minimal quality loss.

Architecture that combines information retrieval from external sources with LLM text generation for more accurate responses.

Technique controlling the number of requests a client can make to an API within a time period, protecting services against abuse and overload.

Database copies dedicated to handling read queries, reducing load on the primary database and improving overall performance.

Process of restructuring existing code without changing its external behavior, improving readability, maintainability, and internal design.

Verify changes do not break functionality

AWS pricing model based on a 1 or 3-year usage commitment that offers significant discounts compared to on-demand pricing.

HTTP interface for system communication

Practice of matching cloud instance sizes to actual workload requirements to optimize costs without sacrificing performance.

Training technique that aligns AI model outputs with human preferences using reinforcement learning.

Financial metric that measures the profitability of an investment by comparing the benefit obtained with the initial cost.

Metric defining the maximum amount of data an organization can afford to lose, measured as the time interval between the last backup and the incident.

Metric defining the maximum acceptable time to restore a service after an incident, determining the urgency and cost of the recovery strategy.

Scalability is the ability of a system to handle growth without degrading performance. Learn the types of scalability and when it matters.

AWS virtual firewall that controls inbound and outbound network traffic at the instance level, defining which connections are allowed through access rules.

Own datacenter where hardware and software are operated in-house vs cloud providers. For sustained AI inference >5B tokens/month, reaches break-even in 4-6 months with 75% savings over 5 years.

AI orchestration SDK developed by Microsoft that enables integrating language models with existing code, plugins, and memory to build enterprise AI applications.

Cloud computing model where the provider manages server infrastructure, scaling, and maintenance automatically.

Infrastructure layer that handles service-to-service communication in microservices architectures, providing observability, traffic management, and security without code changes.

Horizontal database partitioning technique that distributes data across multiple instances to scale storage and query capacity.

Formal agreement between provider and client defining service level commitments, including guaranteed availability, performance, and response times.

Tasks where an AI agent can execute with high confidence and minimal supervision because they have been validated multiple times. Concept from Mitchell Hashimoto framework for effective agent delegation.

Quantitative metric measuring a specific aspect of service level, such as latency or error rate, serving as the foundation for defining SLOs and SLAs.

Internal objective defining the desired reliability level for a service, stricter than the SLA and used as an operational target by engineering teams.

Minimal set of tests verifying that critical application functionalities work correctly after a deployment, before running the full test suite.

Security standard for B2B SaaS

Software architecture is the fundamental structure of a system: components, relationships and design principles. Learn why it is critical for your project success.

SOLID is five principles of object-oriented software design. Learn what each letter means and how to apply them for maintainable code.

Spare AWS compute capacity available at discounts of up to 90% compared to on-demand pricing, with the possibility of interruption.

Fixed-duration development cycle in agile methodologies where the team commits to completing a defined set of work delivering functional increments.

Staff Augmentation is a hiring model where external professionals are added to temporarily reinforce an internal team. Learn when it makes sense.

Architectural design where servers store no session data between requests, enabling effective horizontal scaling and fault tolerance.

Migration pattern that gradually replaces a monolithic system by redirecting functionalities one by one to the new system until the original becomes obsolete.

AI code completion tool that prioritizes privacy and enterprise security, offering models that can run locally without sending code to the cloud.

Total cost of acquiring, implementing, maintaining, and operating a technology solution throughout its entire lifecycle.

Systematic process for identifying, quantifying, and prioritizing technical debt in a system, determining its impact on productivity and remediation cost.

Technical debt is the accumulated cost of suboptimal technical decisions. Learn what it is, types, how to identify it and strategies for managing it.

Parameter that controls the randomness of LLM outputs. Lower values produce deterministic responses; higher values generate more creative, varied outputs.

Benchmark of 89 tasks to evaluate LLM agent capabilities in terminal environments, with realistic tasks from training ML models to compiling Linux from source. Frontier models solve <65% of tasks.

Open-source Infrastructure as Code tool by HashiCorp for defining cloud resources in declarative configuration files with multi-cloud support.

Percentage of code with automated tests

Tiempo desde concepto hasta lanzamiento

Pricing models and cost optimization for LLM usage, where output tokens cost 3-10× more than input tokens, and large requests (>200K) have automatic premium pricing.

Component that splits text into tokens (subword units) that LLMs can process, affecting cost, context limits, and multilingual performance.

LLM capability to invoke external tools like web search, code execution, file operations, and APIs during a conversation.

Decoding strategies that control how an LLM selects the next token, balancing coherence and diversity in generated responses.

Neural network architecture that uses self-attention mechanisms to process sequences in parallel, enabling modern LLMs.

Version control strategy where developers integrate small, frequent changes directly into the main branch, minimizing long-lived feature branches.

Methodology of twelve principles for building modern SaaS applications that are portable, scalable, and deployable on cloud platforms predictably.