Retrieval-augmented generation: search, embeddings, vector stores, knowledge graphs.
Unlike relational databases that compute connections at query time, graph databases persist relationships at insertion time. These relationships are materialized as first-class citizens, allowing queries to simply follow pointers for much faster retrieval.
▶ 0:32
The Graph Layer Behind NASA’s Breakthroughs
Memory enables agents to be reflective, interactive, proactive, and autonomous, mimicking human intelligence where recall determines capability. Without it, applications remain stateless and heavily reliant on repetitive prompt engineering.
▶ 3:15
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Slack combines search and AI to handle massive enterprise data without hitting token limits. Search retrieves relevant raw messages while the LLM synthesizes them with citations, balancing recall and summarization.
▶ 10:51
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Sentry uses embeddings to group errors like anonymous functions with random integers, replacing tedious regular expression rules. This lowers total work and achieves much closer matching compared to historical methods.
▶ 5:44
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
You can use Zerobus, which is an API that handles high-rate, bursty, or tiny-row data ingestion directly into your open lake without requiring a traditional message bus or buffering. Additionally, Spark Real-time Mode enables low-latency streaming operations down to tens of milliseconds.
▶ 17:21
Data + AI Summit Keynote 2026 | Day 1
Self-supervised learning removes the need for expensive manual labels by corrupting raw data and training models to reconstruct it. This approach builds general raw representations that can be fine-tuned for various downstream tasks.
▶ 8:01
Stanford CS25: Transformers United V6 I Overview of Transformers
A language model is essentially a function that predicts the next likely word or token in a sentence based on its training. By predicting successive words, it constructs coherent continuations or fills in missing parts of text. During this process, it evaluates probabilities for various candidate words to determine the most fitting output.
▶ 1:33
A Hackers' Guide to Language Models
Traditional vector search acts as a black box and lacks explainability, meaning you cannot easily tell why specific fragments were retrieved. Furthermore, it often fails to account for broader contextual relationships present in structured or semi-structured data sources.
▶ 8:17
The Graph Layer Behind NASA’s Breakthroughs
Memory management involves generation, storage, retrieval, integration, updating, and specialized deletion mechanisms like forgetting. The retrieval phase is particularly crucial for pulling relevant structured memories into the context window rather than stuffing all data in.
▶ 6:37
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Slack enforces security by running real-time permission checks on search results before passing any data to the LLM. They avoid training models on internal data, relying instead on permissioned context windows.
▶ 14:43
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Seer combines interconnected telemetry data like traces and logs, feeding them into LLMs to perform root cause analysis. Although early versions hallucinated, the model successfully helps diagnose complex production bugs by mimicking the deep theory crafting of senior engineers.
▶ 8:02
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
Apache Iceberg v3 support allows managed Iceberg tables to share data files with Delta tables without requiring any file rewrites. This establishes a unified data layer where data laid out on disk looks identical regardless of the format.
▶ 20:56
Data + AI Summit Keynote 2026 | Day 1
Transformers enable parallel processing of entire sequences on GPUs rather than stepping through them sequentially like RNNs. They also handle much longer contexts and scale up significantly better.
▶ 14:36
Stanford CS25: Transformers United V6 I Overview of Transformers
Language models tokenize text into sub-word units, whole words, punctuation, or numbers using specific tokenizers like tick token. These tokens are mapped to numeric IDs that act as lookups into the model's vocabulary. For instance, prefixes like spaces are often encoded directly into the token representation.
▶ 4:45
A Hackers' Guide to Language Models
Modern LLMs possess strong language skills and can be provided with a specific schema and instructions to extract entities, relationships, and attributes from unstructured text. While not completely error-free, this automated extraction is far more efficient than manual human processing.
▶ 13:58
The Graph Layer Behind NASA’s Breakthroughs
MongoDB provides a flexible document data model that adapts to various data structures alongside multiple retrieval capabilities like graph, vector, text, and geospatial queries in one database. This allows developers to store diverse memory types and build custom memory management solutions.
▶ 8:36
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
The team built internal Model Context Protocol integrations to let coding assistants safely query internal databases and analytics. This helps developers navigate historical context and generate correct SQL queries without hallucinating schemas.
▶ 19:11
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Cursor allows developers to incrementally review diffs and seamlessly jump between files within the editor. In contrast, terminal interfaces create high-friction endless logs that make code review difficult. Therefore, human-in-the-loop workflows are much more effective inside a proper IDE.
▶ 26:28
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
Lakebase Postgres allows you to instantly branch petabyte-scale databases in less than a second using a copy-on-write mechanism on lake storage. This allows AI agents to safely experiment, test, and roll back changes without duplicating underlying storage.
▶ 26:00
Data + AI Summit Keynote 2026 | Day 1
Training small models on child-scale data shows positive scaling trends, though the signal is noisy and heavily task-dependent. Quality, semantic diversity, and structured interactions matter much more than raw data volume at these small scales.
▶ 22:24
Stanford CS25: Transformers United V6 I Overview of Transformers
Pre-training forces a deep neural network to guess the next word across vast corpuses like Wikipedia, utilizing stochastic gradient descent to optimize its weights. To succeed at this task, the model must implicitly learn complex real-world concepts, entities, and relationships. This capability essentially acts as a form of information compression.
▶ 7:40
A Hackers' Guide to Language Models
Wrapping infrastructure and cloud APIs in MCP servers allows developers to manage resources directly within their IDE using rich context from their codebase. This approach reduces the cognitive load of memorizing complex CLI commands or navigating clunky cloud provider user interfaces.
▶ 16:48
The Graph Layer Behind NASA’s Breakthroughs
Persona memory helps assign a consistent personality to AI systems to make them more believable and human-like. Developers can model this personality data inside MongoDB to foster better relationships with users.
▶ 10:09
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Teams can use scripts powered by models like Claude Code to incrementally migrate old code bases into modern languages like TypeScript. For instance, Slack converted over 50% of its legacy JavaScript to TypeScript in just six months. The trade-off is ensuring developers share and tune prompts so the agent matches internal coding standards.
▶ 23:29
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Simple server-side architectures make it much easier for both humans and machines to reason about pattern matching. When an API or framework has minimal boilerplate and clear boundaries, models can successfully generate migrations and tests. This return to simplicity reduces the infinite complexity plaguing modern stacks.
▶ 33:06
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
The Unity AI Gateway serves as a single pane of glass to manage all AI spend, models, and MCP tools across an organization. It provides granular observability dashboards, budget limits and rate-limiting down to the individual level, and enforces security and compliance guardrails.
▶ 34:09
Data + AI Summit Keynote 2026 | Day 1
Training small models on two languages simultaneously does not negatively impact performance on a single language. The perplexities remain low and comparable to single-language baselines regardless of how the languages are interleaved.
▶ 27:11
Stanford CS25: Transformers United V6 I Overview of Transformers
While pre-training instills broad world knowledge, raw models are not inherently optimized for answering questions or solving tasks. Instruction tuning feeds the model targeted datasets of questions and responses to shape its behavior for problem-solving. Additional steps like reinforcement learning from human feedback further refine the outputs by aligning them with human preference.
▶ 15:40
A Hackers' Guide to Language Models
A security vulnerability allowed malicious issues added to a public repository to trick an LLM with access to private repositories into leaking sensitive data. The assistant pulled private repo information to solve the issue and created a public pull request containing the leaked data.
▶ 22:11
The Graph Layer Behind NASA’s Breakthroughs
Instead of stuffing all tool schemas into the limited context window, developers can store JSON schemas of tools inside MongoDB. Just before hitting the LLM, the system queries the database to retrieve only the relevant tool via search.
▶ 10:45
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Large AI-generated pull requests, such as a 5,000-line translation, place an immense burden on human reviewers and often lead to superficial reviews. Reviewers tend to run out of cognitive stamina and skip commenting on massive changes. The recommendation is to pace reviews and be mindful of willpower limits, since slow is smooth and smooth is fast.
▶ 25:44
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Organizations need to enable the consumption of AI coding tools without overly restrictive compliance on intellectual property. Removing the fear of source code leaks or model training risks allows engineering teams to immediately adopt valuable assistants. Without this policy shift, developers are forced to pay out of pocket or fall behind.
▶ 37:10
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
Live retrieval loops force agents to perform random walks through massive enterprise data corpora, leading to high token costs and runtimes of 10 to 15 minutes. Furthermore, this ad-hoc search often results in poor accuracy because the agent lacks proper upfront context.
▶ 37:52
Data + AI Summit Keynote 2026 | Day 1
Smaller models benefit significantly more from retrieval-augmented generation, while larger models saturate very quickly because they have already memorized massive amounts of information. There is a minimum pre-training threshold required before a model can effectively use retrieval.
▶ 30:06
Stanford CS25: Transformers United V6 I Overview of Transformers
Developers can improve model outputs by supplying custom instructions that act as prepended text guiding the model's persona and reasoning quality. Explicitly instructing the model to think through background context helps combat its tendency to generate overly confident hallucinations. Additionally, because generation happens sequentially token by token, prompting the model to reason longer increases its computational thoroughness.
▶ 22:21
A Hackers' Guide to Language Models
The core security challenge is that current AI systems do not separate data from instructions, causing the data and control planes to merge. Because LLMs cannot reliably discern between processing data and executable instructions, they are vulnerable to prompt injections embedded in external data.
▶ 23:21
The Graph Layer Behind NASA’s Breakthroughs
When an agentic system executes steps that fail, that failure can be stored in the database as a learning experience. This historical failure data is then pulled into subsequent execution steps to prevent the LLM from repeating the same mistake.
▶ 12:15
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Slack functions as a work operating system comprising interacting primitives like huddles, lists, canvases, and channels with deep native AI integrations. Using platforms like Agent Force, developers can build agents that dynamically generate canvases, create lists, or fetch enterprise search context. This positions Slack not just as an IDE, but as the core environment for running enterprise workflows.
▶ 33:42
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
AI tools deliver clear productivity gains by rapidly generating boilerplate code and building standard APIs or tests based on existing examples. Furthermore, they re-engage experienced builders who get excited about writing code again. This hands-on involvement improves technical judgment across the organization.
▶ 43:22
Transforming Debugging and Root Cause Detection with Sentry | David Cramer
Genie Ontology automatically connects to structured and unstructured sources like the lakehouse, Google Drive, and SharePoint in the background to build a comprehensive knowledge graph. It uses an algorithm called OntoRank, inspired by PageRank, to index snippets and incorporate human-curated semantics.
▶ 39:10
Data + AI Summit Keynote 2026 | Day 1
Chain of thought prompting forces the model to solve complex math or reasoning problems step by step like a human before producing a final answer. This approach unlocks hidden depth within the model's weights and significantly improves performance on complex tasks.
▶ 38:12
Stanford CS25: Transformers United V6 I Overview of Transformers
The OpenAI API is stateless, meaning nothing is stored on the server between requests. To maintain a conversation, you must pass the entire history of messages back in every new API call.
▶ 43:33
A Hackers' Guide to Language Models
Developers should exercise due diligence, use only trusted MCP servers, and sandbox their environments by restricting access to harmless or open-source repositories. For mission-critical workflows, developers should avoid automated execution and rely instead on manual copy-pasting.
▶ 26:42
The Graph Layer Behind NASA’s Breakthroughs
MongoDB acquired Voyage AI to integrate their embedding models and re-rankers directly into MongoDB Atlas. This integration aims to reduce AI hallucinations and eliminate the need for developers to write custom chunking strategies.
▶ 14:09
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
When building features for agents, developers must expose tools as actions and design search APIs that deliver rich context rather than just single messages. For example, Slack's real-time search API provides full threads, relationships, and user roles so agents can reason effectively. This requires building dedicated endpoints alongside human-facing interfaces.
▶ 47:43
How Slack AI Agents Accelerate Dev Productivity | Samuel Messing
Genie 1 connects to multiple enterprise data sources, including email, calendar, Salesforce, and Databricks, to deliver comprehensive customer profiles. In benchmarks on novel enterprise questions, coding agents only succeeded half the time and hit time limits, whereas Genie 1 provided grounded and accurate results.
▶ 1:02:45
Data + AI Summit Keynote 2026 | Day 1
Hallucinations occur when a model's internally learned world model contradicts an external ground-truth reference world model under a specific evaluation view and conflict policy. This structured mathematical framework successfully separates factual beliefs from agent planning mistakes.
▶ 58:48
Stanford CS25: Transformers United V6 I Overview of Transformers
New or free users often hit low request-per-minute limits initially. You can handle this programmatically by writing a retry function that catches rate limit errors, reads the retry duration, and sleeps before calling itself again.
▶ 44:53
A Hackers' Guide to Language Models
Developers should build their own custom MCP servers to gain a firsthand understanding of how tools, prompts, and resources operate under the hood. Building a simple server takes only about 200 lines of code and helps remove the black-box magic that hinders trust.
▶ 30:58
The Graph Layer Behind NASA’s Breakthroughs
Historical neuroscience research on the visual cortex of cats inspired convolutional neural networks for object and face detection. Today, agent architects collaborate directly with neuroscientists studying human memory to apply nature's blueprint to advanced artificial intelligence systems.
▶ 15:59
Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB
Genie Ontology extracts expressions, relationships, and expertise from underlying data assets, running them through an onto-rank algorithm to inject authoritative context. This approach yields a consistent 30 percentage point improvement in accuracy while cutting runtime in half compared to leading agents.
▶ 1:07:58
Data + AI Summit Keynote 2026 | Day 1
Current language models operate primarily statelessly or via inference-time context windows without updating their core neural weights. Achieving lifelong learning requires permanent, in-weight modifications that can propagate changes to dependent facts without breaking other knowledge.
▶ 1:03:59
Stanford CS25: Transformers United V6 I Overview of Transformers
You pass a list of available tools or functions under the functions keyword argument in the chat completion request. Because you cannot pass raw Python functions directly, you must convert them into JSON schemas using libraries like Pydantic and Python's inspect module.
▶ 47:12
A Hackers' Guide to Language Models
PepsiCo deployed Genie within its supply chain AI brain platform called Spend Wise to handle direct and indirect procurement insights. Within the first few weeks of launch, users drove roughly 30,000 queries, shifting away from traditional dashboards toward conversational data engagement.
▶ 1:28:54
Data + AI Summit Keynote 2026 | Day 1
State-space models process sequences using continuous state updates and compressed internal states analogous to recurrent neural networks. They scale linearly with the number of tokens instead of quadratically, making them much more efficient for long sequences and contexts.
▶ 1:14:42
Stanford CS25: Transformers United V6 I Overview of Transformers
Quantized formats like GPTQ optimize models to use lower-precision data formats. This significantly reduces the memory moving around, making inference faster and allowing larger models to run efficiently on standard hardware.
▶ 1:05:30
A Hackers' Guide to Language Models
LakeFlow replaces legacy Spark with open declarative pipelines that unify batch and streaming workloads using both SQL and Python. Because the frameworks are open and declarative, developers can run them anywhere, while enabling AI agents to easily write the code.
▶ 1:34:38
Data + AI Summit Keynote 2026 | Day 1
Vision transformers break images down into patches and encode them into smaller sequences rather than applying local spatial filters across individual pixels. This avoids inductive biases and allows the model to learn better from massive datasets.
▶ 46:34
Stanford CS25: Transformers United V6 I Overview of Transformers
Instruction-tuned models rely on specific prefix tokens and formatting rules established during their fine-tuning process. Failing to replicate this exact prompt structure results in degraded model performance or incorrect behavior.
▶ 1:08:12
A Hackers' Guide to Language Models
Zerobus Ingest is a fully managed service that is 100% wire compatible with Kafka, allowing existing producers to land data directly into the lakehouse in open formats. It eliminates small file issues and processes millions of small files at high speeds without requiring complex cluster management.
▶ 1:37:16
Data + AI Summit Keynote 2026 | Day 1
Retrieval augmented generation searches external sources like Wikipedia to find relevant documents for a given query. It then injects this retrieved text directly into the prompt context so the language model can generate an accurate, up-to-date response.
▶ 1:11:52
A Hackers' Guide to Language Models
Traditional lake houses hit a latency wall around one second for stringent workloads, forcing companies to maintain separate serving stacks. Raiden flips traditional database engineering by training a machine learning model on exabytes of execution traces to dynamically dispatch the right algorithms at runtime. This enables Lakehouse RT to deliver millisecond performance and massive concurrency directly on governed lake data.
▶ 1:54:49
Data + AI Summit Keynote 2026 | Day 1
You can load a domain-specific dataset using the Hugging Face datasets library and train the model using tools like Axolotl. This process modifies model weights to adapt to specialized structures, such as converting natural language questions into database SQL queries.
▶ 1:22:19
A Hackers' Guide to Language Models
Lakebase decouples storage and compute by moving data storage into the lake while introducing specialized services to maintain performance. Safekeepers handle writes using the Paxos consensus protocol for low-latency transactions, while page servers deliver low-latency reads. This architecture enables fully managed serverless Postgres instances with fast provisioning and scaling.
▶ 2:11:53
Data + AI Summit Keynote 2026 | Day 1
Lakebase offers rapid provisioning in under 500 milliseconds and supports instant branching to create isolated dev, test, and staging environments. Developers and agents can also use snapshot restore via a simple click to test code changes or schema modifications. If an agent's work is unsatisfactory, the database can be instantly rolled back to the previous snapshot.
▶ 2:13:11
Data + AI Summit Keynote 2026 | Day 1
Lakebase provides fully managed cross-cloud disaster recovery to ensure continuous business operations during major cloud outages. Users can configure replication across different cloud providers, such as AWS and Azure. In the event of an outage, the system can instantly fail over to the secondary cloud.
▶ 2:17:32
Data + AI Summit Keynote 2026 | Day 1
Mastercard built a shared foundation using Lakebase to support thousands of issuing banks on a single platform. This architecture ensures strict data isolation and complies with data residency requirements to maintain client trust. Simultaneously, it allows insights to be consolidated securely so agents can reason across data without leaking sensitive multi-tenant details.
▶ 2:19:37
Data + AI Summit Keynote 2026 | Day 1
Mastercard leverages performance pools within Lakebase to maintain strict data separation for thousands of issuing banks while enabling continuous insights. This architecture allows them to scale horizontally while ensuring secure, anonymized data consolidation.
▶ 2:21:39
Data + AI Summit Keynote 2026 | Day 1
Traditional change data capture pipelines read database binary logs to replicate changes, making them fragile and prone to data corruption or maintenance overhead. They often require external infrastructure and can fail under heavy production workloads.
▶ 2:43:32
Data + AI Summit Keynote 2026 | Day 1
LTAP stands for lake transactional analytical processing, which unifies storage by transcoding row-oriented operational data into columnar formats directly on storage services. Unlike traditional failing HTAP systems, it avoids query engine compromises while providing real-time analytics without hurting OLTP performance.
▶ 2:47:55
Data + AI Summit Keynote 2026 | Day 1
Storage services use their underutilized CPUs to transcode row data into columnar Delta and Iceberg formats during local disk writes. Because columnar formats offer high compression ratios, this conversion minimizes data volume and leverages I/O bound resources efficiently.
▶ 2:46:02
Data + AI Summit Keynote 2026 | Day 1
Developers do not need to set up custom pipelines or change their operational workflows to use LTAP. Once enabled, Lakebase tables automatically appear in Unity Catalog in Delta and Iceberg formats for real-time querying.
▶ 2:49:25
Data + AI Summit Keynote 2026 | Day 1
No answer mentions that.