38 things we've learned about building AI products
Practical, no-nonsense insights from building AI products.
Library
Ideas I return to.
A personal collection of work that has shaped my professional thinking on building products, engineering systems, and making decisions.
Selected for their impact.Articles, talks, tools, and research.
Worth keeping. Worth sharing.
336 references
-
-
The Sales Playbook For Founders | Startup School
Tom Blomfield’s playbook for early B2B sales.
-
Postgres Message Queue (PGMQ)
Message queues, directly in Postgres. Also supported by Supabase.
-
Checklist Design
A useful checklist of product details, extending well beyond visual design.
-
Detecting and countering misuse of AI: September 2026
Specific, documented examples of AI misuse and how they were addressed.
-
The Tech Industry Is Huge - and Europe’s Share of It Is Very Small
Jina AI’s path from a framework to selling models for search.
-
The Bravest Defense Plan in Europe
Leaving room in a defence plan for technology that does not exist yet.
-
Hidden Technical Debt in Machine Learning Systems
The long-term engineering costs around a machine learning model.
-
Data Flywheels for LLM Applications
Turn production examples into better evaluations and better applications.
-
What We’ve Learned From A Year of Building with LLMs
Practical lessons across the technical and organizational sides of LLM delivery.
-
Efficient Memory Management for Large Language Model Serving with PagedAttention
The memory-management ideas behind vLLM and PagedAttention.
-
RAG vs. GraphRAG: A Systematic Evaluation and Key Insights
Evidence for choosing between text retrieval and graph-based retrieval.
-
Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
What model-based evaluation can measure, and where it can mislead.
-
ContextCite: Attributing Model Generation to Context
Trace generated statements back to the context that influenced them.
-
Your AI Product Needs Evals
Build evaluation around the errors that appear in real product use.
-
Opening up ChatGPT: tracking openness of instruction-tuned LLMs
A multidimensional view of what “open” means for an AI model.
-
Model Cards for Model Reporting
Document a model’s capabilities, limitations and intended uses.
-
A Practical Guide to Implementing DeepSearch/DeepResearch
How to build a search agent that iterates, checks its work and knows when to stop.
-
Transformers, the tech behind LLMs | Deep Learning Chapter 5
A visual explanation of the transformer architecture behind language models.
-
Introduction to Information Retrieval
The foundations of search, from inverted indexes to retrieval evaluation.
-
Designing Data-Intensive Applications
Understand the trade-offs behind reliable, scalable data systems.
-
LLM Inference Performance Engineering: Best Practices
Measure and improve the latency, throughput and cost of serving LLMs.
-
Emerging Resilience in the Semiconductor Supply Chain
How industrial policy and new investment could change semiconductor dependencies.
-
Chip War: The Fight for the World’s Most Critical Technology
The industrial and geopolitical history behind today’s computing infrastructure.
-
Easily Train a Specialized LLM: PEFT, LoRA, QLoRA, LLaMA-Adapter, and More
A practical introduction to parameter-efficient fine-tuning methods, including LoRA and QLoRA.
-
Foundation Model Course at the University of Amsterdam
University teaching material on foundation models and how to adapt them.
-
NVIDIA H100 GPU
Specifications and architectural context for NVIDIA’s H100 accelerator.
-
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
The BERT paper introduces bidirectional pre-training for language understanding.
-
Improving your LLMs with RLHF on Amazon SageMaker
A worked implementation of reinforcement learning from human feedback using SageMaker.
-
Tune PaLM text models by using RLHF tuning
Historical Google Cloud documentation for tuning PaLM text models with reinforcement learning from human feedback. Covers prompt and human-preference datasets, reward models, tuning jobs, and deployment of tuned models. The original page is no longer available.
-
LLM API Pricing Calculator
A calculator for comparing language-model API costs.
-
ELIZA
An overview of ELIZA and its place in the history of conversational software.
-
Long short-term memory
Background on recurrent networks designed to preserve information across long sequences.
-
aubio/aubio
An audio-analysis library with practical examples of open-source signal processing.
-
zylon-ai/private-gpt
A repository for building AI applications with private data and local models.
-
Gradient Blog: RAG 101 for Enterprise
An introduction to retrieval-augmented generation for enterprise applications.
-
bigscience/xP3all
A multilingual collection of prompted datasets for instruction tuning.
-
What are agents? 🤔
An introduction to agents, their components, and tool use.
-
Text generation
Guidance for generating text with the Transformers library.
-
mistralai/Mistral-Small-24B-Instruct-2501
The model card for Mistral Small 24B Instruct 2501.
-
Ethical Charter - a Hugging Face Space by bigscience
The ethical charter created for the BigScience research collaboration.
-
EcoLogits Calculator
A calculator for estimating the environmental impacts of generative-AI usage.
-
How People Create and Destroy Value with Generative AI
Research into how generative AI changes knowledge-work performance.
-
Aligning language models to follow instructions
OpenAI’s explanation of instruction-following models trained using human feedback.
-
Stanford CoreNLP
Stanford’s toolkit for annotating and analyzing text with natural-language processing methods.
-
Amazon and Anthropic announce strategic collaboration to advance generative AI
The announcement of Amazon and Anthropic’s AI collaboration.
-
La BnF en chiffres
Institutional statistics from the Bibliothèque nationale de France.
-
IA : cette célèbre comédienne américaine poursuit Meta et OpenAI pour violation de droits d'auteur
Reporting on a copyright dispute involving generative-AI training data.
-
D-Day : la famille du vétéran Léon Gautier dit stop au business mémoriel
Reporting on objections to commercial uses of the memory of D-Day veteran Léon Gautier.
-
Large Language Models 101: History, Evolution and Future
An introduction to the development and historical context of large language models.
-
Europarl Parallel Corpus
A parallel corpus built from European Parliament proceedings.
-
The Battle Over Books3 Could Change AI Forever
Reporting on the Books3 dataset and disputes about books used for AI training.
-
[1hr Talk] Intro to Large Language Models
Andrej Karpathy’s introductory lecture on large language models.
-
Knowledge Graph - Deep Learning: A Case Study in Question Answering in Aviation Safety Domain
A case study combining knowledge graphs and deep learning for aviation-safety questions.
-
The 70% problem: Hard truths about AI-assisted coding
The remaining engineering work behind apparently near-complete AI-generated code.
-
AI Essentials For Tech Executives
Teaching material on AI systems and decisions for technology leaders.
-
Introducing Meta Llama 3: The most capable openly available LLM to date
Meta’s introduction to its Llama 3 model family.
-
MLOps
Apache Airflow examples for coordinating machine-learning workflows.
-
Qonto: How Qonto Achieved 70% Faster Localization and Streamlined Operations with AI Agents
Qonto’s account of using AI agents in localization and operational workflows.
-
vLLM V1: A Major Upgrade to vLLM's Core Architecture
The architectural changes introduced with vLLM V1.
-
#15 Le ML Ops pour accélérer l’innovation, de la R&D à la production by Post Mortem
A discussion of MLOps and the path from research to production.
-
ILSVRC 2014 human image-classification interface
An interactive tool for comparing human image classification with ImageNet model predictions.
-
Carrying out error analysis
Lecture slides on inspecting model errors to prioritize machine-learning improvements.
-
TruLens - Pinecone Docs
Documentation on evaluating retrieval applications with TruLens and Pinecone.
-
OpenAI-Compatible Server
Documentation for serving language models through an OpenAI-compatible API with vLLM.
-
Mean reciprocal rank
An explanation of mean reciprocal rank as a retrieval metric.
-
Prompting Fundamentals and How to Apply them Effectively
Prompting techniques grounded in how language models respond to instructions and examples.
-
apple-aiml-research/ml-ferret
Apple’s research code for the Ferret multimodal model.
-
open-webui/open-webui
A user interface for working with language models through local or hosted services.
-
Gen AI at work: The story of Carrefour Marketing Studio - Horizons by Carrefour
Carrefour’s account of developing a generative-AI studio for marketing teams.
-
mistralai/Mistral-Small-3.1-24B-Instruct-2503
The model card for Mistral Small 3.1 24B Instruct 2503.
-
Building A Generative AI Platform
Architectural choices and components involved in building a generative-AI platform.
-
RAG is more than just embedding search
Why retrieval systems need more than a vector search over embeddings.
-
Stop using LGTM@Few as a metric (Better RAG)
Why inspecting only a few retrieval results is a weak evaluation method.
-
What I learned from competing against a ConvNet on ImageNet
Lessons from comparing human image recognition with a convolutional neural network.
-
Extrinsic Hallucinations in LLMs
A survey of hallucination mechanisms and mitigation approaches in language models.
-
LM Studio Bionic - Your Agent for Work and Code
Software and documentation for working with locally hosted language models.
-
Ahead of AI | Sebastian Raschka, PhD | Substack
Sebastian Raschka’s writing on machine learning and AI research.
-
How we built an AI factory: Part 1
How Preligens structured the frameworks, teams, and production practices behind its AI factory.
-
Switching perspectives on generative artificial intelligence: a design view for humans-generative AI co-creativity
A design perspective on collaboration between humans and generative AI.
-
Mistral Small 3.1 | Mistral AI
Mistral’s announcement of the Small 3.1 model.
-
Ollama · Run AI models locally and in the cloud
A runtime for working with language models locally.
-
Hello GPT-4o
OpenAI’s introduction to GPT-4o and its multimodal capabilities.
-
OpenTelemetry
The open observability project covering traces, metrics, and logs.
-
LLM Evals and Benchmarking – hackerllama
An introduction to language-model evaluation and benchmarking.
-
Speed Matters
Research into how web latency affects user behavior.
-
Topic extraction with Non-negative Matrix Factorization and Latent Dirichlet Allocation
A worked example of extracting topics with NMF and latent Dirichlet allocation.
-
LLM Price Comparison
A comparison of language-model price and performance.
-
Imitation Intelligence, my keynote for PyCon US 2024
Simon Willison’s PyCon keynote on large language models and practical software development.
-
WebSockets vs. Server-Sent events/EventSource
A technical discussion comparing bidirectional sockets with one-way browser event streams.
-
Intel Announces Launch of Enterprise Gen AI Spinoff Articul8
Reporting on Intel’s enterprise generative-AI spinout, Articul8.
-
Autonomy in Weapon Systems: DoD Directive 3000.09
The US Department of Defense directive on autonomy in weapon systems.
-
Please Stop Saying "Human-In-The-Loop"
An argument for more precise language about human responsibility in autonomous systems.
-
Changes in OpenAI API pricing
A comparison of changes in OpenAI API pricing.
-
Web Operations: Keeping the Data on Time
Operational practices for running web systems, from monitoring and deployment to incident analysis.
-
Éclairage sur…n°6 – ChatGPT ou la percée des modèles d’IA conversationnels | PEReN
A public technical explainer on ChatGPT and conversational AI models.
-
LLM Evaluation doesn't need to be complicated
A practical introduction to evaluating language models.
-
Understanding and Mitigating Model Drift in Machine Learning
An overview of detecting and responding to model drift.
-
Safran buys AI firm Preligens for 220 million euros
Reporting on Safran’s acquisition of Preligens and its integration as Safran.AI.
-
Calculating GPU memory for serving LLMs
A practical estimate of GPU memory needs from model size and numeric precision.
-
Patent Landscape Report: Generative Artificial Intelligence
WIPO’s analysis of the patent landscape around generative AI.
-
Carrying Out Error Analysis (C3W2L01)
Andrew Ng’s lesson on using error analysis to guide machine-learning work.
-
REX IA Générative avec @CarrefourFrance #iagenerative
A discussion of Carrefour’s experience applying generative AI.
-
Efficient Estimation of Word Representations in Vector Space
The word2vec paper on efficient learning of distributed word representations.
-
Learning Transferable Visual Models From Natural Language Supervision
The CLIP paper on learning visual representations from paired image and text data.
-
Text and Code Embeddings by Contrastive Pre-Training
Research on learning representations for text and code using contrastive training.
-
ChatGPT as a Factual Inconsistency Evaluator for Text Summarization
An evaluation of ChatGPT for detecting factual inconsistencies in summaries.
-
GPT4Graph: Can Large Language Models Understand Graph Structured Data ? An Empirical Evaluation and Benchmarking
An empirical study of how language models handle graph-structured information.
-
Evaluating Correctness and Faithfulness of Instruction-Following Models for Question Answering
Methods for assessing correctness and faithfulness in question answering.
-
Large Language Models for Information Retrieval: A Survey
A survey of language-model applications in information retrieval.
-
RAGAS: Automated Evaluation of Retrieval Augmented Generation
The Ragas paper on evaluating retrieval-augmented generation systems.
-
FlattenQuant: Breaking Through the Inference Compute-bound for Large Language Models with Per-tensor Quantization
A quantization method designed to improve language-model inference efficiency.
-
SOAR: Improved Indexing for Approximate Nearest Neighbor Search
An indexing approach for approximate nearest-neighbor search.
-
From Local to Global: A Graph RAG Approach to Query-Focused Summarization
The GraphRAG paper on graph-based retrieval for broad, query-focused summaries.
-
Recent advances in text embedding: A Comprehensive Review of Top-Performing Methods on the MTEB Benchmark
A review of text-embedding methods evaluated on MTEB.
-
FineSurE: Fine-grained Summarization Evaluation using LLMs
A fine-grained approach to evaluating summaries with language models.
-
ColPali: Efficient Document Retrieval with Vision Language Models
A retrieval method that uses vision-language representations of document pages.
-
Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates
A systematic study of language models used as judges in alignment evaluation.
-
AIR-Bench: Automated Heterogeneous Information Retrieval Benchmark
A benchmark for evaluating retrieval across heterogeneous tasks and domains.
-
Evaluate Summarization in Fine-Granularity: Auto Evaluation with LLM
A study of detailed, language-model-based summarization evaluation.
-
SurveyX: Academic Survey Automation via Large Language Models
A system for automating the preparation and writing of academic surveys.
-
ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback
A method for refining summaries through feedback and reflective reasoning.
-
Evaluation of Approximate nearest neighbors: large datasets
Public vector datasets for benchmarking approximate nearest-neighbor search.
-
Opening up ChatGPT: Tracking openness, transparency, and accountability in instruction-tuned text generators
Research tracking openness and accountability in instruction-tuned language models.
-
Langfuse: Open Source Agent Evals & Observability
An open platform for tracing and evaluating language-model applications.
-
Finding the Right LLM for Your Business: A Practical Cost-Benefit Analysis
A cost-benefit framework for selecting a language model for a business.
-
ACUEval: Fine-grained Hallucination Evaluation and Correction for Abstractive Summarization
A method that decomposes summaries into atomic claims to evaluate hallucinations.
-
ROUGE: A Package for Automatic Evaluation of Summaries
The original description of ROUGE as a package for evaluating summaries.
-
V-JEPA: The next step toward advanced machine intelligence
Meta’s introduction to learning video representations with V-JEPA.
-
Amplifying | Coding agent intelligence
A study of which developer tools Claude Code selects across different tasks.
-
LoRA: Low-Rank Adaptation of Large Language Models
The LoRA paper on adapting large models through low-rank parameter updates.
-
Human-like Summarization Evaluation with ChatGPT
Research into using ChatGPT to evaluate summaries in relation to human judgments.
-
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
A method for learning from preferences without a separate reinforcement-learning reward model.
-
Lost in the Middle: How Language Models Use Long Contexts
Research showing how the location of information affects its use in long contexts.
-
End-to-End Graph Flattening Method for Large Language Models
A method for representing graph information in a form language models can consume.
-
Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking
A study of how presentation style can bias language-model judges.
-
From Single to Multi: How LLMs Hallucinate in Multi-Document Summarization
An investigation of hallucinations when language models summarize multiple documents.
-
Failure Modes of LLMs for Causal Reasoning on Narratives
A study of failures in causal reasoning about narrative information.
-
Injecting Knowledge Graphs into Large Language Models
Research on bringing structured knowledge graphs into language models.
-
Forewarned is Forearmed: A Survey on Large Language Model-based Agents in Autonomous Cyberattacks
A survey of autonomous cyberattacks carried out using language-model agents.
-
Device-Edge Cooperative Fine-Tuning of Foundation Models as a 6G Service
Research on sharing foundation-model fine-tuning between devices and edge infrastructure.
-
Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs
The HNSW paper describing graph-based approximate nearest-neighbor search.
-
Attention Is All You Need
The Transformer paper introducing an architecture built around attention.
-
Generate synthetic data for evaluating RAG systems using Amazon Bedrock
A worked method for generating synthetic evaluation data for RAG systems.
-
Biome
A development tool combining code formatting and linting.
-
Best Practices for LLM Inference Performance Monitoring
A guide to the metrics and operational practices used to monitor language-model inference.
-
vLLM: Easy, Fast, and Cheap LLM Serving with PagedAttention
The original vLLM announcement and its approach to memory-efficient model serving.
-
Configuration as Code (CaC): A DevOps guide - CircleCI
An introduction to managing deployment configuration as code.
-
BFloat16: The secret to high performance on Cloud TPUs
An explanation of the bfloat16 number format used in Cloud TPUs.
-
Mistral AI models | Gemini Enterprise Agent Platform | Google Cloud Documentation
Google Cloud documentation for supported Mistral models.
-
European Chips Act
The European Commission’s overview of the European Chips Act.
-
Semiconductors
A 2020 overview of Taiwan’s semiconductor industry, its manufacturing clusters, policy initiatives, and investment opportunities.
-
ANN Search is super slow
A technical discussion of slow approximate-nearest-neighbor search in Elasticsearch.
-
Create your first Java-based simulation
A tutorial for creating and running a first load test with Gatling.
-
Models Overview
Mistral’s overview of its models and API versions.
-
NVIDIA System Management Interface (nvidia-smi)
NVIDIA’s reference for querying and managing GPU state through nvidia-smi.
-
Metrics - Ragas
Reference documentation for the evaluation metrics implemented in Ragas.
-
Prometheus and Grafana
A guide to observing vLLM with Prometheus and Grafana.
-
Tensorize vLLM Model
An example of serializing and loading vLLM models with Tensorizer.
-
Multimodal Inputs
Documentation for sending multimodal inputs to vLLM.
-
Offline Inference - vLLM
Reference documentation for offline inference with vLLM.
-
Usage Stats Collection
The usage data collected by vLLM and the options for disabling collection.
-
Frequently Asked Questions
Frequently asked questions about operating vLLM.
-
Structured Outputs
Documentation for constraining generated outputs to a required structure.
-
Vector database management systems: Fundamental concepts, use-cases, and current challenges
An overview of vector database concepts, use cases, and implementation challenges.
-
Carousel Personalization in Music Streaming Apps with Contextual Bandits
Research on personalizing music-app carousels using contextual bandits.
-
Betweenness centrality
An explanation of betweenness centrality in graph analysis.
-
Binary search
The binary-search algorithm and its computational properties.
-
Confusion matrix
A reference on interpreting classification outcomes with a confusion matrix.
-
Discounted cumulative gain
An explanation of discounted cumulative gain for ranking evaluation.
-
Dog whistle (politics)
Background on coded political messaging and its interpretation.
-
Gartner hype cycle
An explanation of the Gartner hype-cycle model of technology expectations.
-
Likert scale
A reference on Likert scales for collecting and interpreting judgments.
-
Rank (linear algebra)
An explanation of matrix rank in linear algebra.
-
Spearman's rank correlation coefficient
A reference on rank correlation and how it measures monotonic association.
-
Faiss: A library for efficient similarity search
Meta’s introduction to the Faiss library for vector similarity search.
-
AI Engineer 2023 Keynote - Building Blocks for LLM Systems
A keynote on evaluation, retrieval, guardrails, and feedback in LLM systems.
-
Deepfake Macron : Créez aisément une vidéo de Macron par l'IA
An example of consumer software for generating synthetic video of a public figure.
-
Lithographie
Background on lithography and its role in producing patterned surfaces.
-
Quantile
An explanation of quantiles for understanding distributions.
-
bitsandbytes-foundation/bitsandbytes
A library for low-bit model operations and quantization.
-
huggingface/sentence-transformers
A library for training and using embedding and reranking models.
-
confident-ai/deepeval
A framework for evaluating language-model applications.
-
Guidelines to choose an index
Faiss guidance on choosing an index for similarity search.
-
k-quants by ikawrakow · Pull Request #1684 · ggml-org/llama.cpp
The llama.cpp pull request introducing the k-quants quantization methods.
-
grafana/k6
An open-source load-testing tool built around JavaScript and Go.
-
huggingface/text-generation-inference
Hugging Face’s server for text-generation model inference.
-
huggingface/trl
Tools for training language models with reinforcement learning and preference data.
-
langfuse/langfuse
The source repository for Langfuse’s tracing and evaluation platform.
-
locustio/locust
A load-testing tool that defines user behavior in Python.
-
Model Context Protocol
The open protocol connecting AI applications to tools and data sources.
-
pgvector/pgvector
A PostgreSQL extension for storing vectors and searching by similarity.
-
state-spaces/mamba
Research code for the Mamba state-space architecture.
-
traceloop/openllmetry
OpenTelemetry-based instrumentation for generative-AI applications.
-
ultralytics/ultralytics
The Ultralytics computer-vision repository.
-
unslothai/unsloth
Tools for training and running language models with reduced resource requirements.
-
vllm-project/vllm
The source repository for vLLM’s model-serving engine.
-
Full-stack observability for the agentic era | Grafana Labs
The Grafana observability platform and its technical resources.
-
A Framework for Offline Evaluation of Recommender Systems based on Probabilistic Relational Models
A framework for evaluating recommender systems that use relational data.
-
Object Storage - Utiliser Object Storage avec S3cmd - OVHcloud Documentation
OVHcloud instructions for configuring S3cmd with object storage.
-
NousResearch/Hermes-4-70B
The model card for Nous Research’s Hermes 4 70B.
-
Qwen/Qwen2.5-7B-Instruct-1M
The model card for Qwen2.5 7B Instruct with a million-token context window.
-
TheBloke/CodeLlama-7B-GGUF
A GGUF distribution of Code Llama 7B with quantized variants.
-
Making LLMs even more accessible with bitsandbytes, 4-bit quantization and QLoRA
A guide to 4-bit model loading and QLoRA with bitsandbytes.
-
Train 400x faster Static Embedding Models with Sentence Transformers
A method for efficiently training static text embeddings.
-
Training and Finetuning Embedding Models with Sentence Transformers
Guidance for training and adapting Sentence Transformers embedding models.
-
HuggingFaceH4/instruction-dataset
An instruction-tuning dataset hosted by Hugging Face H4.
-
ChaoticNeutrals/Creative_Writing-ShareGPT
A ShareGPT-format creative-writing dataset used as an example of fine-tuning data.
-
sentence-transformers/gooaq
The GooAQ question-answering dataset in Sentence Transformers format.
-
sentence-transformers/msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1
A Sentence Transformers training dataset derived from MS MARCO.
-
Downloading models
Instructions for downloading model files from Hugging Face.
-
Command Line Interface (CLI)
Reference documentation for the Hugging Face command-line interface.
-
PEFT
Documentation for the PEFT parameter-efficient fine-tuning library.
-
Text Generation Inference
Documentation for Hugging Face Text Generation Inference.
-
Supported Models
The model architectures supported by Text Generation Inference.
-
TRL - Transformers Reinforcement Learning
Documentation for the TRL reinforcement-learning and preference-training library.
-
DPO Trainer
Reference documentation for the direct preference optimization trainer.
-
Migrating from OpenAI to Open LLMs Using TGI’s Messages API
A guide to using open models through TGI’s Messages API.
-
meta-llama/Llama-2-7b-chat-hf
The model card for Meta’s Llama 2 7B chat model.
-
meta-llama/Llama-3.2-1B
The model card for Meta’s Llama 3.2 1B model.
-
mistralai/Mistral-Small-3.1-24B-Base-2503
The base-model card for Mistral Small 3.1 24B.
-
mistralai/Mistral-Small-3.1-24B-Instruct-2503 · Address discrepancies in the languages supported by the Mistral Small 3.1 2503
A discussion about inconsistencies in the documented languages supported by Mistral Small 3.1.
-
MTEB Leaderboard - a Hugging Face Space by mteb
A leaderboard for comparing text-embedding models across MTEB evaluations.
-
stabilityai/stable-diffusion-xl-base-1.0
The model card for Stable Diffusion XL 1.0.
-
Product Quantization for Nearest Neighbor Search
Research on compact vector representations for approximate nearest-neighbor search.
-
How much does AI cost in 2025? Well, it depends
An overview of the factors that determine the cost of developing AI applications.
-
JanusGraph · Distributed, open source, massively scalable graph database
A graph database and its technical documentation.
-
Declutter your JavaScript & TypeScript projects
A tool for finding unused files, dependencies, and exports in JavaScript projects.
-
Observability Data Model - Langfuse
The data model used to represent traces and observations in Langfuse.
-
Welcome to Apache Lucene
The Apache Lucene search library and its documentation.
-
BERTopic
A topic-modeling library that combines embeddings with class-based TF-IDF.
-
Practical Tips for Finetuning LLMs Using LoRA (Low-Rank Adaptation)
Practical observations from fine-tuning experiments with LoRA.
-
Makefile Tutorial by Example
An example-driven introduction to Makefiles.
-
Unveiling the Dynamics of Vector Databases in AI landscape
An introduction to vector storage, semantic search, and criteria for selecting a vector database.
-
Reverse Engineering the Semiconductor Industry
An introduction to chip-industry history, manufacturing models, and supply-chain dependencies.
-
Transformers KV Caching Explained
An illustrated explanation of reusing attention states to speed up autoregressive generation.
-
OpenAI GPT-3 Text Embeddings - Really a new state-of-the-art in dense text embeddings?
An analysis of OpenAI’s text-embedding models and their evaluation.
-
Leiden Clustering for Community Detection: A Step-by-Step Guide with Python Implementation
A Python-based guide to identifying graph communities with the Leiden algorithm.
-
What is Milvus?
An introduction to Milvus and its architecture for storing and searching vector data.
-
Fine-tune Mistral-7b with Direct Preference Optimization
A worked guide to fine-tuning Mistral 7B with direct preference optimization.
-
Quantize Llama models with GGUF and llama.cpp
A guide to quantizing models for use with llama.cpp.
-
louvain_communities
NetworkX documentation for Louvain community detection.
-
pagerank
NetworkX documentation for computing PageRank.
-
OpenLLM observability integration
An observability integration for monitoring OpenLLM.
-
Introducing deep research
OpenAI’s introduction to its research-oriented agent.
-
European Open Source AI Index
An index assessing the openness of AI systems.
-
DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node
A method for billion-point approximate nearest-neighbor search on one machine.
-
pre-commit
A framework for managing and sharing pre-commit hooks.
-
Prometheus - Monitoring system & time series database
An open monitoring system and time-series database.
-
Filterable HNSW Without Recall Loss
An explanation of filtered vector search using HNSW.
-
Filtering
Qdrant documentation for constraining search results with filters.
-
Swiss Cheese Model – SAFEChE: Process Safety
An explanation of layered defenses through the Swiss cheese model.
-
Loss Overview
A guide to choosing loss functions for Sentence Transformers.
-
scikit-learn metrics reference
Scikit-learn’s reference for scoring functions and evaluation metrics.
-
LatentDirichletAllocation
Reference documentation for latent Dirichlet allocation in scikit-learn.
-
ShareGPT-4o
A dataset of multimodal annotations generated with GPT-4o.
-
Music recommendation at Spotify
A presentation about the research and systems behind music recommendations at Spotify.
-
Cost Optimized hosting of Fine-tuned LLMs in Production | Microsoft Community Hub
A deployment pattern for reducing the cost of serving fine-tuned models.
-
RAG evaluation metrics and frameworks
A discussion of the metrics and frameworks available for evaluating retrieval-augmented generation.
-
How to leverage large language models without breaking the bank
Inference costs and open models.
-
How TSMC Surpassed Intel in Advanced Chip Technology
An account of TSMC’s rise in advanced chip manufacturing.
-
FACT SHEET: CHIPS and Science Act Will Lower Costs, Create Jobs, Strengthen Supply Chains, and Counter China | The White House
The White House fact sheet introducing the CHIPS and Science Act.
-
Speech and Language Processing: RNNs and LSTMs
A textbook chapter explaining recurrent neural networks and long short-term memory models.
-
Parameter Efficient Tuning
Course slides introducing methods that adapt language models with relatively few trainable parameters.
-
ASML EUV lithography systems
ASML’s explanation of extreme-ultraviolet lithography systems.
-
What in the world is FinOps, and why do we need it? - Inside Atlassian
An introduction to FinOps and the management of cloud expenditure.
-
GenAI Needs Pricing Strategies to Match Its Potential
A pricing framework connecting customer value, differentiation, and the changing costs of generative AI.
-
7 ways to deploy your own large language model
A comparison of approaches to deploying language models.
-
ChatGPT and generative AI are booming, but the costs can be extraordinary
Reporting on the infrastructure costs behind generative AI.
-
L’espace indopacifique : une priorité pour la France | France Diplomatie
France’s diplomatic priorities and regional partnerships in the Indo-Pacific.
-
Improved Text Scoring with BM25
A technical presentation on BM25 relevance scoring in Elasticsearch.
-
Beyond Benchmarks 2024
A study of AI-model performance beyond conventional benchmark scores.
-
Your Success with Generative AI May Come Down to These UX Decisions
Product-interface decisions that affect the usefulness and adoption of generative AI.
-
The Economics of LLMs
A comparison of cost factors in selecting and deploying language models.
-
What Large Models Cost You: There Is No Free AI Lunch
An account of the financial and operational costs of building and running language models.
-
What is a context window?
An explanation of context windows and their constraints in language models.
-
What is a LLM pricing strategy?
An overview of pricing approaches for language-model services.
-
What Is a Corporate Spin-Off?
An explanation of corporate spin-offs and their financial structure.
-
Searching in One Billion Vectors: Re-rank with Source Coding
A vector-search method that refines nearest-neighbor results using compact quantization codes.
-
The Diminishing Returns of Deploying Local Large Language Models
An analysis of diminishing returns in deploying language models locally.
-
LLM Economics: Which is Cheaper to deploy ChatGPT Vs Open Source LLMs?
A cost comparison between hosted commercial models and self-hosted open models.
-
How to Choose the Best Embedding Model for Your LLM Application
Guidance on choosing an embedding model for an application that uses language models.
-
User intent and implicit feedback in AI agents: what every enterprise leader needs to know - Nebuly - The ROI platform for Enterprise AI
An explanation of explicit and implicit user feedback for AI products.
-
Counting The Cost Of Training Large Language Models
An analysis of the cost of training large language models.
-
NVIDIA A100 Tensor Core GPU datasheet
NVIDIA’s technical datasheet for the A100 Tensor Core GPU.
-
NVIDIA L40 datasheet
NVIDIA’s technical datasheet for the L40 GPU.
-
Hierarchical Navigable Small Worlds (HNSW) | Pinecone
An illustrated explanation of HNSW vector-search indexes.
-
7 Top Large Language Model Use Cases And Applications
Examples of business applications for large language models.
-
Using Object Storage with s3cmd
A guide to configuring s3cmd for Scaleway Object Storage. Covers regional credentials and configuration files, bucket and object operations, and CORS configuration.
-
Google "We Have No Moat, And Neither Does OpenAI"
A discussion of open models and competitive defensibility in AI.
-
The Inference Cost Of Search Disruption – Large Language Model Cost Analysis
An analysis of the inference economics of language-model-based search.
-
US Semiconductor Ecosystem Map
A map of the US semiconductor ecosystem.
-
Fine-tuning large language models (LLMs) in 2026
A guide to adapting large language models through fine-tuning.
-
The $360K question about Large Language Models Economics
An analysis of costs and tradeoffs in running large language models.
-
TruLens: Evals and Tracing for AI Agents
Documentation and examples for tracing and evaluating AI applications with TruLens.
-
Feedback Functions
Introduces programmatic feedback functions for evaluating LLM application runs. Compares expert judgments, user feedback, traditional NLP metrics, and model-based evaluations by scalability and usefulness.
-
The RAG Triad
An explanation of groundedness, relevance, and context quality in RAG evaluation.
-
La start-up française Mistral AI lève 105 millions d'euros pour développer des modèles d'IA générative
Reporting on Mistral AI’s early fundraising and plans to develop generative-AI models.
-
L’Allemagne espère rebondir après l’abandon par Intel de son projet de méga-usine à 30 milliards d'euros
Reporting on Intel’s cancelled German fabrication plant and the implications for industrial policy.
-
Normal Distribution Percentile: Formula & Graph | Vaia
A worked explanation of percentiles in a normal distribution.
-
RDF 1.1 Primer
The W3C primer introducing the RDF 1.1 data model.
-
NVIDIA Blackwell: The Journey From Die to Data Center
A visual explanation of the journey from NVIDIA’s Blackwell chips to a data center.
-
What Makes Large Language Models Expensive?
An explanation of the factors behind the cost of large language models.
-
How are Microchips Made? 🖥️🛠️ CPU Manufacturing Process Steps
An introduction to the manufacturing steps used to produce microchips.
-
Bases de Données Vectorielles : Expérience & Conseils d'Expert
A discussion of vector databases with practical deployment experience.
-
SurveyX and automated academic surveys
A technical thread explaining how SurveyX organizes references and generates research surveys.
-
Using reasoning models differently from chat models
A reminder that reasoning models need different usage patterns from conventional chat models.
-
Comparing open language models
A comparison figure updated to include DeepSeek-V2 and Mistral Large 2.
-
tinyML Talks: A Practical Guide to Neural Network Quantization
A technical talk introducing neural-network quantization.
-
Accelerating LLM Inference with vLLM
A presentation by vLLM contributors on accelerating language-model inference.
-
A Survey of Sustainability in Large Language Models: Applications, Economics, and Challenges
A survey of environmental, economic, and computational challenges in sustainable language models.
-
The Environmental Impacts of AI: Policy Primer
A policy primer on the resources and emissions involved across the lifecycle of AI systems.
-
Open Container Initiative Charter
The governance charter defining the mission and operation of the Open Container Initiative.
-
Bylaws | Linux Foundation
The bylaws governing membership, leadership, and decision-making at the Linux Foundation.
-
Apple Releases Swift as Open Source
Apple’s announcement that Swift became an open-source programming language.
-
How to become an amazing contributor (to an open source project)
Practical expectations for becoming an effective contributor to an open-source project.
-
From Openness to Transparency: The Role of Social Media in Open Source Ecosystems
A keynote on collaboration, coordination, and visibility in open-source ecosystems.
-
The Changing Open Source Landscape
A reflection on how the incentives and culture of open source have changed.
-
Government Transparency via Open Data and Open Source
An argument for building public accountability through open data and open-source software.
-
Building Leadership in an Open Source Community
Guidance for developing trust and leadership within open-source communities.
-
Voting for transparency
Reporting on open-source voting infrastructure and public confidence in elections.
-
10 ways to contribute to an open source project without writing code
Ways to support open-source projects beyond writing code.
-
Ageism in tech: the not-so-invisible age limit developers face
An examination of age-related barriers in software-development careers.
-
What to consider with an open source business model
A discussion of the decisions behind commercial open-source business models.
-
Transparency and Democracy in open source: not what you thought
An examination of how open-source projects disclose their governance and decision-making.
-
Folding and Unfolding: Balancing Openness and Transparency in Open Source Communities
Research on how open-source developers balance visible collaboration with protected working spaces.
-
How working transparently united our engineers
An account of how transparency helped engineers at Buffer collaborate.
-
Open Source and Transparency: Not The Same Thing
An explanation of the distinction between sharing source code and operating transparently.
-
A Primer on ASF Governance | Apache Software Foundation
The Apache Software Foundation’s corporate governance structure and procedures.
-
Open Source Transparency
A discussion of incentives, openness, and accountability in software communities.
-
opengovernance/opengovernance.dev
A checklist for evaluating ownership and decision-making in open-source governance.
-
Open source: Security through transparency | IoT Now News & Reports
A discussion of the security implications of transparent software development.
-
Software Transparency
An examination of software transparency as a defense against hidden vulnerabilities.
-
The real benefits of open: Why Adjust lives by open-source programming
Adjust explains the reasoning behind its open-source software and SDK practices.
-
European Chips Survey
The European Commission’s survey of semiconductor demand and supply-chain constraints.
-
La stratégie indopacifique de la France
France’s strategic priorities in the Indo-Pacific, including security, regional cooperation, and economic dependencies.
No matching references.
Try a broader idea, another source, or fewer filters.
Every entry links to the original work and a citation you can reuse. Notes explain what stood out to me; related references help connect ideas across formats.