| 1 | Activation FunctionWIP A non-linear function applied element-wise to a neuron's output so a stack of layers can represent more than a single linear map. | Term | Deep Learning |
| 2 | Agent LoopWIP The control loop an LLM agent runs — call the model, execute any tool calls it returns, feed the results back, repeat until it stops asking for tools. | Term | Agents |
| 3 | Agentic RAGWIP A RAG system where an LLM agent actively decides what to retrieve, evaluates retrieval quality, and iterates on its search strategy rather than doing a single retrieve-and-generate pass. | Term | Agents |
| 4 | Apache KafkaWIP You need durable, high-throughput event streaming with multiple independent consumers replaying history. | Tool | Streaming |
| 5 | AttentionWIP A mechanism that lets a model weigh which other tokens in a sequence matter most when computing a representation for the current one. | Term | Deep Learning |
| 6 | AutoGenWIP You want agents that converse with each other to solve problems, especially when code generation and execution are part of the workflow. | Tool | LLM & Agent Frameworks |
| 7 | BackpropagationWIP The algorithm that computes gradients of the loss with respect to every weight in a network by applying the chain rule backwards. | Term | Deep Learning |
| 8 | Batch NormalizationWIP A layer that normalizes activations across a mini-batch so each feature has roughly zero mean and unit variance, then rescales them with learned parameters. | Term | Deep Learning |
| 9 | Beam SearchWIP A decoding strategy that keeps the top-k highest-probability partial sequences at each step instead of greedily picking one. | Term | LLMs |
| 10 | BenchmarkWIP A standardized test suite with fixed tasks and metrics used to compare model capabilities across a consistent yardstick. | Term | Eval |
| 11 | BentoMLWIP You have a trained model and need to ship it as a reliable, scalable API with proper batching, runner management, and Docker packaging. | Tool | API & Serving |
| 12 | BiasWIP The statistical error introduced when a model is too simple to capture the true structure of the data — not the societal fairness kind. | Term | Classical ML |
| 13 | Canary DeploymentWIP Rolling out a new model version to a small percentage of traffic first, monitoring key metrics, and only promoting to full traffic if things look good. | Term | MLOps |
| 14 | Chain of ThoughtWIP Prompting an LLM to write out its reasoning step by step before giving a final answer, which usually improves accuracy on multi-step problems. | Term | LLMs |
| 15 | ChromaWIP Prototyping, local development, or small production apps where a lightweight embedded store is enough. | Tool | Vector Databases |
| 16 | ChunkingWIP Splitting documents into smaller, retrievable pieces before embedding them so retrieval returns the right span instead of the wrong book. | Term | RAG |
| 17 | Concept DriftWIP When the relationship between inputs and the target variable changes over time, meaning the patterns your model learned are no longer correct -- even if the input distribution stays the same. | Term | MLOps |
| 18 | Context WindowWIP The maximum number of tokens an LLM can consider in a single forward pass — prompt plus generated output. | Term | LLMs |
| 19 | Convolutional Neural NetworkWIP A neural network architecture that uses learned convolution filters over local regions of an input grid — images, audio spectrograms, or any tensor with spatial structure. | Term | Vision |
| 20 | CrewAIWIP You want multiple LLM agents collaborating on a task with defined roles, and you prefer a high-level declarative API over wiring graph nodes. | Tool | LLM & Agent Frameworks |
| 21 | Cross-ValidationWIP A resampling technique that estimates how well a model will generalize by training and evaluating on multiple data splits. | Term | Classical ML |
| 22 | Data DriftWIP When the distribution of incoming data shifts away from the distribution the model was trained on, causing predictions to degrade even though the model itself hasn't changed. | Term | MLOps |
| 23 | DatabricksWIP You're running Spark-scale ETL, training models on terabytes, and want notebook + job + MLflow in one place. | Tool | Data Warehouses |
| 24 | Decoder-OnlyWIP A transformer architecture that uses only the decoder stack with causal masking, treating every task as next-token prediction on a single concatenated sequence. | Term | LLMs |
| 25 | DeepEvalWIP You want to write automated eval suites for your RAG pipeline or LLM features and run them in CI. | Tool | MLOps |
| 26 | Diffusion ModelWIP A generative model that learns to reverse a gradual noising process — you train it to denoise, then sample by iteratively denoising pure noise. | Term | Deep Learning |
| 27 | Direct Preference OptimizationWIP A simpler alternative to RLHF that skips the reward model entirely and directly optimizes the language model on preference pairs using a classification-style loss. | Term | LLMs |
| 28 | DockerWIP You need reproducible environments for training, serving, or local development — especially anything involving CUDA, Python versions, or system libraries. | Tool | Infrastructure |
| 29 | DriftWIP When the data or relationships a model sees in production move away from what it was trained on, quietly degrading performance. | Term | MLOps |
| 30 | DropoutWIP A regularization trick that randomly zeros out a fraction of activations during training so the network can't depend on any single neuron. | Term | Deep Learning |
| 31 | DSPyWIP You have a well-defined task with examples and want the framework to automatically search over prompts, few-shot demos, and even fine-tunes. | Tool | LLM & Agent Frameworks |
| 32 | EmbeddingWIP A dense vector that encodes the meaning of a piece of data so similar things land near each other in vector space. | Term | Deep Learning |
| 33 | EncoderWIP The part of a model that reads an input sequence and produces a fixed-size or per-token vector representation for downstream use. | Term | Deep Learning |
| 34 | Encoder-DecoderWIP An architecture where one stack of layers reads the full input bidirectionally (encoder) and another generates the output autoregressively (decoder), connected by cross-attention. | Term | Deep Learning |
| 35 | EpochWIP One full pass of the training algorithm over the entire training dataset. | Term | Classical ML |
| 36 | F1 ScoreWIP The harmonic mean of precision and recall — a single number that punishes you for being lopsided on either. | Term | Eval |
| 37 | FastAPIWIP You need a typed, async Python HTTP service — especially one that serves ML models, proxies LLM calls, or exposes a RAG pipeline. | Tool | API & Serving |
| 38 | Feature EngineeringWIP The craft of deriving inputs that make a model's job easier — usually by encoding domain knowledge the raw data doesn't surface. | Term | Classical ML |
| 39 | Feature StoreWIP A centralized system for storing, versioning, and serving precomputed features so that training and inference pipelines use the same feature definitions and values. | Term | MLOps |
| 40 | Few-Shot LearningWIP Teaching a model a new task by giving it a handful of example input/output pairs in the prompt rather than by fine-tuning. | Term | LLMs |
| 41 | Fine-TuningWIP Continuing training on a pretrained model with a smaller, task-specific dataset to specialize its behavior. | Term | LLMs |
| 42 | Foundation ModelWIP A large model pre-trained on broad data that serves as the starting point for many downstream tasks via fine-tuning, prompting, or adaptation. | Term | LLMs |
| 43 | Generative Adversarial NetworkWIP Two networks trained against each other — a generator that creates fake samples and a discriminator that tries to tell them from real ones. | Term | Deep Learning |
| 44 | Gradient DescentWIP An iterative optimization method that nudges parameters in the opposite direction of the gradient to reduce a loss function. | Term | Math |
| 45 | GradioWIP You want the fastest path from a Python function to a clickable demo, especially for ML models with image, audio, or text I/O. | Tool | Visualization |
| 46 | GraphRAGWIP A RAG architecture that builds a knowledge graph from your documents and uses graph traversal alongside vector search to retrieve richer, more connected context. | Term | RAG |
| 47 | Greedy DecodingWIP The simplest decoding strategy -- at each step, pick the single token with the highest probability and commit to it. | Term | LLMs |
| 48 | GuardrailsWIP Runtime checks and constraints that prevent an LLM or agent from producing harmful, off-topic, or policy-violating outputs. | Term | Agents |
| 49 | HallucinationWIP When an LLM produces text that is fluent and confident but factually wrong or unsupported by its input. | Term | LLMs |
| 50 | HeliconeWIP You want instant visibility into LLM usage, costs, and latency across your team without instrumenting every call site. | Tool | MLOps |
| 51 | HNSWWIP A graph-based algorithm for approximate nearest neighbor search that builds a multi-layer navigable small-world graph, giving you sub-millisecond vector lookups at scale. | Term | RAG |
| 52 | HuggingFace TransformersWIP Loading, fine-tuning, or running any pretrained transformer model in Python. | Tool | LLM & Agent Frameworks |
| 53 | Hybrid SearchWIP Combining dense vector search (semantic similarity) with sparse keyword search (BM25) to get the best of both retrieval worlds. | Term | RAG |
| 54 | HyperparameterWIP A configuration value you set before training that controls how the model learns — distinct from the parameters the model learns itself. | Term | Classical ML |
| 55 | InferenceWIP Running a trained model to produce predictions on new inputs — the serving side of ML, as opposed to training. | Term | MLOps |
| 56 | JAXWIP A Python library from Google for high-performance numerical computing with autodiff, JIT compilation, and vectorization via a NumPy-like API. | Term | Deep Learning |
| 57 | k-Means ClusteringWIP An unsupervised algorithm that partitions n points into k clusters by iteratively assigning points to the nearest centroid and moving centroids to the mean of their cluster. | Term | Classical ML |
| 58 | KL DivergenceWIP An asymmetric measure of how much one probability distribution differs from another — zero when they match, larger as they diverge. | Term | Math |
| 59 | Knowledge DistillationWIP Training a smaller "student" model to mimic the outputs of a larger "teacher" so you get most of the quality at a fraction of the cost. | Term | Deep Learning |
| 60 | KubernetesWIP You run multiple services at nontrivial scale and need rolling deploys, autoscaling, and a uniform way to manage them. | Tool | Infrastructure |
| 61 | KV CacheWIP Storing the key and value tensors from previous tokens during autoregressive generation so you don't recompute attention over the entire sequence at every step. | Term | LLMs |
| 62 | LanceDBWIP You want vector search without running a server — local development, embedded apps, or datasets that live on S3. | Tool | Vector Databases |
| 63 | LangChainWIP You need a quick LLM application scaffold with ready-made integrations for vector DBs, document loaders, and LLM providers. | Tool | LLM & Agent Frameworks |
| 64 | LangGraphWIP You're building an agent with branching logic, retries, checkpoints, or human-in-the-loop steps and want explicit control over the flow. | Tool | LLM & Agent Frameworks |
| 65 | LangSmithWIP Debugging, evaluating, and monitoring LLM chains or agents in dev and production — especially if you're already on LangChain/LangGraph. | Tool | MLOps |
| 66 | Latent SpaceWIP The lower-dimensional vector space a model compresses its inputs into, where similar things end up near each other. | Term | Deep Learning |
| 67 | Learning RateWIP The scalar that controls how big a step the optimizer takes in the direction of the gradient — the single most important training hyperparameter. | Term | Deep Learning |
| 68 | LlamaIndexWIP You're building a RAG system over a corpus of documents and want ready-made loaders, indexers, and query engines. | Tool | LLM & Agent Frameworks |
| 69 | LLM-as-JudgeWIP Using a language model to evaluate the quality of another model's output, replacing or supplementing human evaluation with automated scoring. | Term | Eval |
| 70 | LLMOpsWIP The operational practices for deploying, monitoring, and maintaining LLM-powered applications in production -- MLOps adapted for the specific challenges of large language models. | Term | MLOps |
| 71 | LoRAWIP A parameter-efficient fine-tuning method that freezes the base model and trains small low-rank update matrices on top. | Term | LLMs |
| 72 | Loss FunctionWIP A scalar score that measures how wrong a model's predictions are — the thing the optimizer tries to make smaller. | Term | Deep Learning |
| 73 | Mixture of ExpertsWIP An architecture where only a few specialized sub-networks ("experts") fire for any given input, chosen by a learned router. | Term | Deep Learning |
| 74 | MLflowWIP You need experiment tracking and a model registry without buying a full MLops platform, and you want something self-hostable. | Tool | MLOps |
| 75 | Model Context ProtocolWIP An open protocol for connecting LLM applications to tools, data sources, and prompts in a standard way — like USB for AI tool integrations. | Term | Agents |
| 76 | Model RegistryWIP A centralized catalog that tracks model versions, their metadata (metrics, lineage, training config), and their lifecycle stage (staging, production, archived). | Term | MLOps |
| 77 | MongoDBWIP Your data is document-shaped, your schema evolves, or you want vector search and operational storage in one place. | Tool | Databases |
| 78 | Multi-Agent SystemWIP An architecture where multiple specialized LLM agents collaborate, debate, or delegate tasks to each other, rather than having one model do everything. | Term | Agents |
| 79 | Multi-Head AttentionWIP Running several attention operations in parallel with different learned projections so the model can attend to multiple relationships at once. | Term | Deep Learning |
| 80 | Neural NetworkWIP A model built from stacked layers of linear transformations and non-linear activations, trained end-to-end with gradient descent. | Term | Deep Learning |
| 81 | OllamaWIP You want to run an open LLM on your laptop or a small server with zero setup — demos, prototypes, offline work. | Tool | API & Serving |
| 82 | OptimizerWIP The algorithm that updates model weights from their gradients — SGD, Adam, AdamW, Lion, and friends. | Term | Deep Learning |
| 83 | OverfittingWIP When a model memorizes training data instead of learning patterns that generalize to new data. | Term | Classical ML |
| 84 | Parameter-Efficient Fine-TuningWIP A family of methods that fine-tune only a small fraction of a model's parameters (or add small trainable modules) instead of updating all weights, drastically cutting memory and compute costs. | Term | LLMs |
| 85 | PerplexityWIP An intrinsic LLM metric — the exponentiated average negative log-likelihood the model assigns to held-out text. Lower is better. | Term | Eval |
| 86 | pgvectorWIP You already run Postgres and your vector workload is modest to medium — tens of millions of vectors, single-digit ms queries. | Tool | Vector Databases |
| 87 | PineconeWIP You need a production vector store and don't want to operate one yourself. Especially good when you need serverless scaling. | Tool | Vector Databases |
| 88 | Positional EncodingWIP The signal added to token embeddings so the model knows where each token sits in the sequence, since self-attention has no built-in notion of order. | Term | Deep Learning |
| 89 | PostmanWIP You're exploring a third-party API, debugging your own service, or sharing request collections with teammates who don't live in a terminal. | Tool | Infrastructure |
| 90 | Pre-trainingWIP The initial phase where a model learns general representations from massive unlabeled data before being adapted to any specific task. | Term | Deep Learning |
| 91 | Precision and RecallWIP Two complementary metrics for classification — precision is "of what I flagged, how much was right"; recall is "of what was actually there, how much did I catch". | Term | Eval |
| 92 | Prompt InjectionWIP An attack where malicious instructions hidden in untrusted input override the developer's prompt and steer the model into doing something it shouldn't. | Term | LLMs |
| 93 | PydanticWIP Validating or serializing structured data in Python — API payloads, configuration, LLM outputs, anything with a schema. | Tool | Infrastructure |
| 94 | PyTorchWIP Any deep learning work — training from scratch, fine-tuning, research, or deploying custom models. | Tool | Deep Learning Frameworks |
| 95 | QdrantWIP You need a production-grade vector database with rich filtering, payload storage, and real operational maturity. | Tool | Vector Databases |
| 96 | QuantizationWIP Storing model weights and activations in lower precision (int8, int4) so the model is smaller, cheaper, and faster — at some quality cost. | Term | MLOps |
| 97 | RagasWIP You have a RAG pipeline and need quantitative metrics beyond eyeballing outputs, especially for regression testing and comparison. | Tool | MLOps |
| 98 | RAGASWIP An evaluation framework specifically designed for RAG pipelines, measuring retrieval quality and generation quality through metrics like faithfulness, answer relevancy, and context precision. | Term | Eval |
| 99 | Ray ServeWIP You're serving multiple models or a complex inference pipeline and need independent scaling, composition, and cluster-level resource management. | Tool | API & Serving |
| 100 | Re-rankingWIP A second-stage retrieval step where a more expensive model re-scores the top-k candidates from the initial retrieval to push the most relevant results to the top. | Term | RAG |
| 101 | ReActWIP An agent pattern that interleaves reasoning ("think") with tool calls ("act") so the model can observe results and adjust mid-task. | Term | Agents |
| 102 | Red-TeamingWIP Systematically probing an LLM or AI system with adversarial inputs to find failure modes, safety violations, and vulnerabilities before real users do. | Term | Eval |
| 103 | RedisWIP You need sub-millisecond access to small hot data — caches, counters, queues, rate limits, session state. | Tool | Cache / In-memory |
| 104 | ReflectionWIP A pattern where an LLM agent reviews its own output, identifies flaws, and revises -- critique-then-fix in a loop. | Term | Agents |
| 105 | RegularizationWIP Anything you add to training that discourages the model from fitting noise — usually a penalty on weight magnitude or randomness in the forward pass. | Term | Classical ML |
| 106 | RerankerWIP A second-stage model that re-scores the top results from a fast retriever to push the most relevant ones to the top. | Term | RAG |
| 107 | Retrieval-Augmented GenerationWIP A pattern where you retrieve relevant documents at query time and stuff them into the LLM's prompt so it can ground its answer in real sources. | Term | RAG |
| 108 | RLHFWIP Fine-tuning a language model using a reward model trained on human preference data, with reinforcement learning to optimize for the reward. | Term | LLMs |
| 109 | ROC-AUCWIP The area under the ROC curve — a threshold-independent measure of how well a classifier ranks positives above negatives. | Term | Eval |
| 110 | scikit-learnWIP Tabular data, prototyping, feature engineering pipelines, and almost any classical ML baseline. | Tool | Deep Learning Frameworks |
| 111 | Self-AttentionWIP The mechanism where every token in a sequence computes attention scores against every other token in the same sequence to figure out which parts of the input matter to which other parts. | Term | Deep Learning |
| 112 | Similarity SearchWIP Finding the items in a database whose vector representations are closest to a query vector, typically using cosine similarity or dot product in an embedding space. | Term | RAG |
| 113 | SnowflakeWIP You want a no-ops analytics warehouse with strong governance, concurrent BI workloads, and simple SQL semantics. | Tool | Data Warehouses |
| 114 | SoftmaxWIP A function that turns a vector of real numbers into a probability distribution — exponentiate each, divide by the sum. | Term | Math |
| 115 | Speculative DecodingWIP A latency optimization where a small draft model generates several candidate tokens quickly, and the large target model verifies them in a single forward pass. | Term | LLMs |
| 116 | StreamlitWIP You want a quick internal tool or demo around a model and don't want to touch frontend code. | Tool | Visualization |
| 117 | Supervised Fine-TuningWIP Training a pre-trained language model on curated (prompt, response) pairs so it learns to follow instructions and produce the kind of output you want. | Term | LLMs |
| 118 | System PromptWIP The hidden instruction block at the start of a conversation that sets the model's persona, constraints, and behavior before the user says anything. | Term | LLMs |
| 119 | TemperatureWIP A scalar that divides logits before softmax at sampling time — lower temperature makes the model more deterministic, higher makes it more random. | Term | LLMs |
| 120 | TensorFlowWIP You're maintaining an existing TF codebase, targeting mobile/edge via TFLite, or serving models through TF Serving at scale. | Tool | Deep Learning Frameworks |
| 121 | TerraformWIP You're managing cloud infrastructure (VPCs, instances, databases, IAM) and want versioned, reviewable, repeatable deployments. | Tool | Infrastructure |
| 122 | TokenWIP The atomic unit an LLM actually reads and writes — usually a sub-word fragment, not a whole word. | Term | LLMs |
| 123 | TokenizerWIP The component that converts text to and from the integer token IDs an LLM actually consumes. | Term | LLMs |
| 124 | Top-K SamplingWIP A decoding strategy that restricts the next-token choice to only the K most probable tokens, then samples from that truncated distribution. | Term | LLMs |
| 125 | Top-P SamplingWIP A decoding strategy that dynamically selects the smallest set of tokens whose cumulative probability exceeds P, then samples from that set. | Term | LLMs |
| 126 | Transfer LearningWIP Taking a model pretrained on a large general dataset and reusing it — with or without further training — on a different, usually smaller, task. | Term | Deep Learning |
| 127 | TransformerWIP A neural network architecture built on stacked self-attention and feed-forward layers — the backbone of every modern LLM. | Term | Deep Learning |
| 128 | UnderfittingWIP When a model is too simple or trained too little to capture the structure in the data — high error on both training and test sets. | Term | Classical ML |
| 129 | Vanishing GradientWIP When gradients shrink exponentially as they propagate back through many layers, so early layers barely update and the network won't train. | Term | Deep Learning |
| 130 | Vector DatabaseWIP A database whose primary index is built for fast approximate nearest-neighbour search over high-dimensional embeddings. | Term | RAG |
| 131 | vLLMWIP You're self-hosting an open-weights LLM and care about throughput, latency, and GPU utilization. | Tool | API & Serving |
| 132 | WeaviateWIP You want an open-source, self-hostable vector store with hybrid (vector + keyword) search out of the box. | Tool | Vector Databases |
| 133 | Weights & BiasesWIP You're training models and want painless experiment tracking with great visualizations, team sharing, and artifact versioning. | Tool | MLOps |
| 134 | XGBoostWIP Any supervised learning problem on tabular data — especially classification, regression, and ranking. | Tool | Deep Learning Frameworks |
| 135 | XGBoostWIP A gradient-boosted decision tree library known for winning tabular-data competitions and still being the right default for structured data. | Term | Classical ML |
| 136 | YOLOWIP A family of real-time object detection models that predict bounding boxes and class scores in a single forward pass. | Term | Vision |
| 137 | Zero-Shot LearningWIP Getting a model to perform a task it wasn't explicitly trained on, purely from a natural-language description of the task. | Term | LLMs |
