Loading…
Browse by type
Explainer reads
Explainer posts on Yomu, summarised newest first.
- 1,318 posts
- 20 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Explainer posts
Evaluation-First AI Agents: How Zepto Scales Customer Support on Databricks and MLflow
Zepto describes its customer-support system, processing more than 100,000 tickets daily, and the assurance problems that emerged as volume, categories, and customer behavior changed. Its response is an evaluation-first architecture on Databricks and MLflow, using development and production loops connected by a quality gate. Agent traces capture prompts, completions, retrieved documents, tool calls, latencies, and decision paths through MLflow, OpenTelemetry spans, and Unity Catalog-backed Delta tables; stakeholders define evaluation pillars and gates around a living golden dataset. The dataset grew from 500 examples with an eight-point development–production accuracy gap to 5,247 examples with a 0.4-point gap over six months, while prompt optimization and multiple scorers support iteration. Reported outcomes include 80% plus of tickets fully managed with human oversight, a 65% reduction in support cost or tickets, 20% higher CSAT, 8% better accuracy, and faster development and resolution.
The Open Source AI Stack
The post presents an open model AI stack for developers moving from closed-source models, focusing on ownership, control, and economics. It organizes the stack into independent layers: model, inference, gateways and routers, harness, and tools such as Skills and MCP. Models generate responses from prompts; large models generally handle ambiguity, multi-step reasoning, and long context better, while small models can be faster and cheaper for tightly specified work. The separation lets developers choose providers and models independently and experiment with newly released models without rebuilding their workflow, while the post uses Kimi K3 and GLM 5.3 Flash to illustrate differences in active parameters, capability, speed, and cost.
The Anatomy of Harness Engineering: How to Evaluate, Iterate, and Guard AI Coding Agents
The post argues that end-to-end benchmarks such as Terminal-Bench and DeepSWE show overall agent performance but rarely explain why behavior changes. It presents behavioral evaluations as fast, deterministic, unit-style checks of observable intermediate actions, such as asking clarifying questions, running validators, or invoking web search instead of guessing. The proposed workflow starts with a recent failure, uses strict milestone assertions for simple tasks and flexible outcome-based or LLM-as-a-judge checks for complex tasks, then runs batch evaluations to track aggregate pass rates despite model nondeterminism. These evaluations are positioned for the second phase of development, after an agent can dogfood its codebase, and as complementary to macro benchmarks: behavioral tests enable prompt, tool-schema, and model iteration while guarding against regressions.
How Metadata and Runtime Telemetry Reveal Which Enterprise System Problems to Fix First
Salesforce Health Insights addresses enterprise org health monitoring by combining configuration metadata with runtime telemetry to identify and prioritize remediation. The team evolved a static report into an in-app experience with durable finding dispositions, approximately 400+ signals across domains, and standards informed by Well-Architected principles and implementation anti-patterns. A shared harmonization layer reconciles fragmented datasets, connector patterns link runtime activity to metadata components, and the system preserves data origin, freshness, and access requirements while keeping deterministic findings distinct from externally supplied context. Prioritization combines findings with request volume, application and database CPU, call behavior, and peak-business-hour traffic to rank next best actions by criticality and effort. A headless JSON schema supports interfaces including an MCP connection, while the deterministic core is intended to produce dependable recommendations and extend access beyond the primary interface.
Ceilometer: Uber’s Adaptive Benchmarking Framework
Ceilometer is Uber’s adaptive benchmarking framework for qualifying new server types, cloud SKUs, kernel upgrades, configuration changes, and machine-learning frameworks before production rollout. It replaces manual, fragmented testing with standardized configurations, automated cluster bring-up and service deployment, portable containerized harnesses, and reporting that includes server specifications, reference SLAs, normalization, resource utilization, cost efficiency, and reliability signals. Its distributed architecture runs synthetic and production-like workloads across dedicated machines, processes and enriches raw results, stores them in blob storage and a centralized data warehouse, and exposes them through an analytics service. The framework covers CPU and memory benchmarks, tools such as SpecCPU2017, SPECjbb2015, NetPerf, and FIO, plus stateful, stateless, batch, and ML workloads; Uber reports uses including regression detection, database validation, and canary testing of external upgrades.
SQL Data Types: Reference and Best Practices
SQL data types define what values columns accept, how those values are stored, and how databases process them, making type selection central to schema design. The reference covers numeric, character, date/time, and specialized types, noting that MySQL, PostgreSQL, SQL Server, and Oracle differ in naming, precision, limits, and storage requirements. It recommends choosing the smallest type that safely accommodates expected values, checking range, precision, NULL requirements, string length, and time-zone needs before deployment. Examples contrast CHAR with VARCHAR, DECIMAL with FLOAT, DATE with TIMESTAMP, and integer sizes such as TINYINT, INT, and BIGINT; indexes and distributed engines also benefit from right-sized data. Sample data, execution-plan analysis, storage measurements, schema audits, and documentation help validate choices, identify waste or overflow risks, and preserve reasoning behind edge cases.
Build durable agents with Temporal and Lakebase
Long-running personal-loan underwriting agents must preserve completed work, resume after worker failures, tolerate repeated tool calls, wait for reviewers, expose operational state, and retain an audit trail. The reference implementation combines Temporal durable Workflows and Activities with Lakebase Postgres, using Temporal Event History for replay and Lakebase schemas for run status, evidence, review state, messages, and metrics. Deterministic identifiers, constraints, guarded updates, and Postgres upserts make repeated Activity attempts target the same logical records, while Signals carry human decisions and long waits do not occupy workers. A Unity Catalog synced table supplies governed underwriting policy, and optional Change Data Feed publishes operational changes to Delta history tables. The implementation includes a React/FastAPI interface, mocked deterministic providers, OAuth credential refresh, and 21 passing tests, but does not validate production lending, compliance, security, scale, or an observed end-to-end feed.
Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic
Topic-level safety systems can conflate a dangerous subject with dangerous intent, causing models to refuse legitimate requests alongside harmful ones. The paper studies deployment-specific boundaries using political persuasion as a testbed: refuse manipulation or targeted persuasion while answering factual political questions. Its training pipeline repairs self-generation coverage through escalating retries, adds verified surface-dangerous benign examples, and evaluates held-out harmful-benign prompt pairs rather than relying on aggregate refusal rates. On Qwen3-8B, escalated coverage raises in-distribution political refusal from 9.47% to 84.75%, while the strongest configuration reduces unsafe responses across three broader benchmarks from 26.26% to 0.14%; however, XSTest over-refusal rises to 74.00%. Adding benign boundary data reduces over-refusal on comply-worthy held-out prompts from 32.94% to 4.16%, with harmful-side refusal declining from 91.88% to 87.72%.
Why Spotify Is Not Using Bayesian A/B Testing
Spotify argues that Bayesian A/B testing is not a single method but a family of configurations defined by a prior, likelihood, and stopping rule. It contrasts flat-prior posterior-probability stopping, which can reproduce frequentist peeking false-positive rates, with Bayes-factor stopping, which controls false positives under optional stopping, and calibrated empirical Bayes priors, which can shrink estimates and control false discovery rates. The post frames experiment design around program goals such as limiting harmful or ineffective launches, achieving estimation precision, and minimizing costs rather than selecting a statistical framework first. It says Bayesian and frequentist approaches overlap substantially, while Bayesian advantages such as richer probability statements require additional prior complexity. Spotify therefore continues using frequentist tooling because supporting two modes would add planning, monitoring, interpretation, and configuration costs without a sufficient benefit.
Debug Past an HTTP 403 Without Breaking Spring Security
Spring Security can reject a request before execution reaches a controller breakpoint, leaving developers to determine the effective roles and whether bypassing authorization is safe for debugging. IntelliJ IDEA’s security inlays read the live SecurityFilterChain during an active debug session, show role or authority requirements, link to configuration, and let developers unlock an exact URI and HTTP method either with all listed roles or with a chosen username and authority set. The debugger populates an artificial Authentication in the running application’s SecurityContext without changing configuration, while CSRF checks and method-level authorization still apply separately. Unlocks last for the debug session or until relocked, affect every client reaching that route, and have limitations including Servlet-only support, no programmatic API, and a null @AuthenticationPrincipal result.
Social Media Management & Consumer Intelligence | Brandwatch
Brandwatch describes an all-in-one social media suite that combines AI, analytics, channel management, consumer intelligence, and specialized tools for enterprise teams. Its consumer-intelligence offering provides access to the world’s largest archive of consumer opinion and uses proprietary and generative AI to identify emerging trends, while threat monitoring covers more than 100 million sources with real-time alerts. A collaborative content calendar supports channels, teams, workflows, approvals, and campaigns, and a unified social inbox handles community management and customer support across networks. The stated goal is to help organizations make better decisions, align content with their brands, respond quickly to threats, and produce measurable results and return on investment; the page also cites recognition by industry analysts and customer trust.
The 40-year-old database rule agents just broke: How LTAP unifies OLTP and OLAP workloads
LTAP (Lake Transactional/Analytical Processing) addresses the long-standing separation between operational row-oriented databases and analytical column-oriented systems, which traditionally required pipelines and copied data. It uses Lakebase’s stateless, ephemeral compute decoupled from durable lake storage to expose live operational data for analytical queries without moving it or loading the transaction-serving system. The storage layer preserves Postgres data types and physical representation while writing data into Parquet, with a hot row-format tier for operational access and a cooler columnar tier for analytical reads; Apache Spark and SQL can read that representation directly. By separating serverless operational and analytical compute, LTAP allows each workload to use specialized engines and independent capacity, while a unified catalog applies governance across both. The stated motivation is agentic workloads such as fraud detection, where stale copies miss fast events and uncontrolled analytical queries can degrade transactions.
What if Stripe could work the way your business does
Stripe’s default financial infrastructure covers common payment, subscription, invoicing, and money-movement needs, but businesses may require custom rules, data models, or operational views. Rather than placing every customization behind webhooks and a separately managed server, Stripe extensibility moves logic, data, and user interface capabilities into Stripe through Scripts, Workflows, Custom Objects, and UI extensions. Scripts handle individual decisions such as TypeScript-based proration, while Workflows orchestrate event-driven or manually initiated sequences involving Stripe APIs, third-party integrations, and custom actions. Custom Objects add typed fields, validation, and relationships for concepts such as Rocket Rides’ Vehicles and Rentals, and UI extensions expose that information and business actions in the Dashboard. In the example, an End Rental action updates status, triggers inspection, and marks a vehicle available, connecting the three dimensions without requiring a separate operational system.
Achieving Extreme Efficiency through Specialized GPU Kernel Generation
Production inference systems often use generic GPU kernels even though operation shapes combine static model parameters with request-time dimensions such as token counts. Proteus addresses this mismatch by having agents propose specialized kernels, then applying controlled-reference validation, static checks, isolated repeated timing, hidden tests, and consistency checks before retaining measured winners. Its knowledge layer uses actionable, scoped lessons retrieved through hierarchical tags and hybrid keyword-semantic search, while background jobs handle deeper distillation instead of burdening each attempt. On NVIDIA B200 GPUs with a Triton backend, a Gated DeltaNet packed-decode case for Qwen 3.5 122B produced shape-specific wins, including 1.6× for a Batch=4, Key=128, Value=128 path and 0.018 ms latency for Candidate 030. Across the reported kernels, Proteus achieved 1.8–5.2× speedups over the best available in vLLM, with validation and context management identified as the main difficulties.
Five ways marketers can use Genie One
Marketing teams often have data across campaign, customer, attribution, and pipeline systems but still depend on analysts to answer questions quickly and consistently. Databricks Genie One is presented as a data-smart AI coworker that uses governed business data, organizational context, metric definitions, campaign structures, reporting rules, and access permissions to support marketing analysis. The five described uses cover channel and campaign performance, funnel attribution, customer insights, omnichannel planning, and mobile business reporting, with natural-language follow-ups, reusable skills or Genie Agents, scheduled analyses, dashboards, and shareable outputs. Examples include tracing sourced pipeline changes, identifying declining enterprise-account usage, and finding territories with weakening repeat purchases. The source recommends starting with repeatable, cross-source workflows and preparing data sources, attribution models, fiscal-calendar logic, and team terminology in Databricks; marketers retain responsibility for decisions and campaign plans.
Pipelined SQL in ClickHouse 26.8
ClickHouse 26.8 introduces pipelined SQL, extending the earlier ability to place FROM before SELECT with a |> operator for sequencing query transformations. Using the UK property prices dataset, the post compares conventional, FROM-first, and pipelined forms for filtering London sales, aggregating median prices by district, ordering results, and limiting output. Each pipeline stage acts as a complete-query checkpoint, while EXPLAIN SYNTAX shows that ClickHouse translates the syntax into nested standard SQL and optimizes the complete query without materializing every intermediate result. The syntax also supports EXTEND, reuse of aggregation aliases in later stages, views, and INSERT ... SELECT statements, though stage order affects which rows feed subsequent aggregations in these examples; the conclusion is that pipelines complement rather than replace conventional SQL.
What is bring your own LLM (BYO LLM)?
Bring your own LLM (BYO LLM) is an architecture in which teams choose the model powering their AI agents while a platform supplies communication infrastructure. The division leaves model selection, prompts, reasoning logic, tool definitions, and called systems with the team, while channels, conversation context, audio streaming, identity resolution, and human escalation remain with the platform. The post explains why bundled models create quality, migration, compliance, and cost-control constraints, including the inability to route simple interactions to smaller models and complex work to frontier models. It describes adding a third-party model to an existing contact center through middleware, a shared message handler, tools and knowledge, channel registration, and context-preserving handoff. Twilio Agent Connect is presented as a self-hosted Python and TypeScript SDK under the MIT License, with voice streaming and integrations for existing contact-center workflows.
How to Carry User Identity Across Federated Kubernetes and AI Platforms
Federated data and AI platforms let users move among portals, datasets, notebooks, and assistants across clusters, but single sign-on alone does not carry trusted identity across control-plane and data-plane boundaries. The post presents a central identity gateway that owns platform sessions through OIDC, a shared session store, stateless data-plane gateways, and an identity-validation API. Regional gateways validate sessions through /gateway/userinfo, enforce local policy, and inject standardized identity headers instead of forwarding raw tokens or independently managing refresh and logout. This separates session ownership from request enforcement, supports consistent logout, and lets assistants pass delegated user identity to backend tools without broad service credentials. At NVIDIA, the approach reduced repeated login events by 55% across internal developer platforms spanning Kubernetes clusters in AWS and OCI, while establishing a foundation for unified platform shells and AI assistants.
Governance beyond security: knowledge, context & ontology on the lakehouse
The post argues that healthcare AI governance must extend beyond security controls to include knowledge, context, and ontology: security governs access, while semantics determine what data means and whether models can trust it. DEP frames classification tags, model cards, data contracts, and lineage as machine-readable governance artifacts, with Unity Catalog serving as the metadata runtime for agents. Build agents can generate ETL, testing, de-identification, and deployment, while analytic agents answer questions over individual data products; human stewards approve proposals and own accountability. A shared lifecycle certifies the data product and AI track through gates using scores for governance, quality, semantics, and ownership, with certification expiring after relevant changes or failures. The approach applies data-layer ABAC to SQL and vector retrieval, logs agent prompts with lineage, and connects readiness metrics to model economics by enabling smaller models when cataloged data supplies sufficient context.
Building High-Quality and Trusted Data Products with Databricks
Enterprise data products are presented as managed assets that deliver the right data at the right time, with appropriate quality, format, and usability for users and applications. Their value depends on clear ownership, business-aware data product owners, and five trustworthiness characteristics: quality and observability, semantic consistency, privacy, security, and discoverability. The proposed lifecycle runs from inception and design through creation, publication, operation and governance, consumption, and retirement, with data contracts, quality metrics, testing, versioning, access management, and lineage supporting those stages. For Databricks implementations, the guidance recommends Delta Live Tables expectations and constraints, staged development, testing and production environments, automated monitoring, and Unity Catalog for discovery, governance, lineage, auditing, and access controls.
Top topics
- Architecture830
- Performance812
- LLMs648
- Machine Learning617
- Open Source561
- AI Agents542
- AI454
- Developer Experience446
Weekly digest
The best engineering reads, once a week.