---
title: "Architecture"
description: "834 posts about Architecture, summarised, each linking to the original."
---

# Architecture
> 834 posts about Architecture, summarised, each linking to the original.

## Articles

### [Connecting Agents to Decisions](https://yomu.fyi/post/connecting-agents-to-decisions.md)
- Company: [Palantir](https://yomu.fyi/company/palantir.md)
- Author: Palantir
- Published: Apr 28, 2026
- Type: Explainer

Palantir’s Ontology is presented as a decision-centric architecture for human-agent operations, intended to represent how enterprises make and execute decisions rather than only storing or analyzing data. It combines data, logic, action, and security in a shared, dynamic representation that can connect operational data, computational processes, workflow execution, and policy controls. The data layer includes structured, streaming, edge, unstructured, imagery, and decision data, with decision lineage recording the data version, application, and decision context; logic binding connects heterogeneous business, modeling, planning, and optimization assets for human and agent workflows. The architecture exposes approved actions to agents with configurable human review and guardrails, while capturing decisions as learning material for improving workflows and agentic memory. It concludes that a real-time ontology can anchor AI in enterprise conditions while controlling where automation is allowed.


### [From Clicks to Conversions: Architecting Shopping Conversion Candidate Generation at Pinterest](https://yomu.fyi/post/from-clicks-to-conversions-architecting-shopping-conversion-candidate.md)
- Company: [Pinterest](https://yomu.fyi/company/pinterest.md)
- Author: Pinterest Engineering
- Published: Apr 27, 2026
- Type: Problem & solution

Pinterest built a dedicated shopping candidate generation model to optimize for lower-funnel purchase conversions rather than relying strictly on engagement retrieval models. Because offsite conversion events are sparse and delayed, engineers trained a single multi-surface two-tower model using conversion data supplemented by duration-weighted click engagement and unengaged ad impressions as hard negatives. The two-tower retrieval architecture incorporates a parallel DCN v2 and MLP layer structure that allows simultaneous learning from the same inputs without information bottlenecks. Later revisions transitioned the system from a multi-head loss structure to a unified single-head multi-task architecture paired with an advertiser-level objective. Deployed across more than 600 million monthly active users, the system increased shopping conversion volume by 2.3% and boosted return on ad spend by 3.1% for United States shopping campaigns.


### [ClickHouse at Google Cloud Next '26](https://yomu.fyi/post/clickhouse-at-google-cloud-next-26.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Apr 27, 2026
- Type: Announcement

ClickHouse announced several product updates and integrations designed to deepen native compatibility with Google Cloud. Migration of ClickHouse Cloud to Google Axion ARM-based processors demonstrated a 30 to 55 percent query speed increase and reduced data load times from 64 to 26 seconds on ClickBench benchmarks. An integration with Google Lakehouse Runtime Catalog enables querying Iceberg tables on Google Cloud Storage via the DataLakeCatalog engine without data movement. Furthermore, Bring Your Own Cloud reached general availability on Google Cloud Marketplace, and ClickPipes added an unordered mode using Pub/Sub notifications for GCS ingestion. Google Antigravity integrated ClickHouse through an MCP server to translate natural language into SQL and generate interactive React charts.


### [How to build scalable web apps with OpenAI's Privacy Filter](https://yomu.fyi/post/how-to-build-scalable-web-apps-with-openai-s-privacy-filter.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: yuvraj sharma, Freddy Boulton, Abubakar Abid
- Published: Apr 27, 2026
- Type: Tutorial

The article presents three web applications built around OpenAI’s open-source Privacy Filter, a PII detector that labels eight categories in one forward pass over a 128k-token context. Document Privacy Explorer extracts PDF or DOCX text and returns category-labeled spans, while Image Anonymizer combines Tesseract word boxes with character offsets to produce editable pixel redaction rectangles. SmartRedact Paste replaces detected spans with category placeholders, stores public and token-gated views, and evicts expired pastes in a daemon thread. Across the apps, gradio.Server’s @server.api routes model work through Gradio’s queue and gradio\_client, while plain FastAPI routes serve custom HTML, views, and lightweight lookups. The implementations keep browser-side filtering, canvas edits, and PNG export out of server round-trips, and the paste service is about 200 lines including storage.


### [DeepSeek-V4: a million-token context that agents can actually use](https://yomu.fyi/post/deepseek-v4-a-million-token-context-that-agents-can-actually-use.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: ben burtenshaw
- Published: Apr 24, 2026
- Type: Explainer

DeepSeek V4 is presented as an open model family aimed at long-running agentic workloads, with Pro and Flash MoE checkpoints offering 1M-token context windows. Its central architecture alternates Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), compressing KV entries by 4x and 128x respectively, while using FP8/BF16 storage and FP4 indexing. The post-training design preserves reasoning across user turns when tool calls are present, introduces a dedicated XML tool-call schema, and relies on DSec, a Rust sandbox platform for RL rollouts. At 1M tokens, Pro uses 27% of V3.2's single-token FLOPs and 10% of its KV cache, while Flash uses 10% and 7%; V4-Pro-Max reports 80.6% on SWE Verified and 0.59 MRCR 8-needle retrieval at 1M.


### [How We Built a Smarter Pickup Experience for Gated Communities](https://yomu.fyi/post/how-we-built-a-smarter-pickup-experience-for-gated-communities.md)
- Company: [Lyft](https://yomu.fyi/company/lyft.md)
- Author: winnieyan
- Published: Apr 23, 2026
- Type: Problem & solution

Pickups in gated communities account for up to 30% of Lyft rides in certain markets, yet locked gates and lack of advance instructions historically resulted in higher cancellation rates and extended wait times. To resolve these friction points, Lyft's Mapping team developed an end-to-end mapping and routing framework. The team built an algorithm to generate gate area boundaries, enabling the app to detect gated communities and offer riders explicit choices between waiting inside or meeting outside the gate. Routing services incorporate the gate as an intermediate waypoint to direct drivers along the proper approach path while triggering timely gate access instructions. Following deployment, rides with shared gate instructions exhibited lower cancellation rates from both parties, shorter walking distances to pickup spots, and fewer extended driver wait times.


### [How to Use Transformers.js in a Chrome Extension](https://yomu.fyi/post/how-to-use-transformers-js-in-a-chrome-extension.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Nico Martin
- Published: Apr 23, 2026
- Type: Tutorial

The guide explains how to build a Chrome extension that runs local AI with Transformers.js under Manifest V3, using a browser assistant as the reference architecture. It assigns model orchestration, lifecycle management, tool execution, and conversation state to a background service worker, while a side panel handles chat and a content script handles page extraction and highlighting. The implementation uses Gemma 4 E2B in q4f16 for text generation and all-MiniLM-L6-v2 in fp32 for embeddings, with WebGPU, DynamicCache-based KV caching, explicit download progress, and reusable pipelines. Typed message contracts, model-output parsing, and separate internal and UI transcripts support a deterministic tool loop, while chrome.storage.local, IndexedDB, and background caches place data according to lifecycle and access needs; the guide concludes that separating these runtimes keeps the UI responsive and avoids duplicate model loads.


### [Frontend Engineering at Palantir: Engineering Multilingual Collaboration](https://yomu.fyi/post/frontend-engineering-at-palantir-engineering-multilingual-collaboratio.md)
- Company: [Palantir](https://yomu.fyi/company/palantir.md)
- Author: Palantir
- Published: Apr 22, 2026
- Type: Explainer

Palantir describes the frontend work behind adding read-only translation mode to Dossier, a collaborative rich-text editor used in multilingual, restrictive environments where public translation services may be unavailable or disallowed. The design lets users choose source and target languages, view a complete document in translation, copy text, and add comments while routing requests through the in-house Language Model Service. Engineers extracted shared translation components, added a Redux state slice and per-element cache, chunked content by paragraphs and other blocks, and used Intersection Observer, batching, debouncing, and caching to translate visible content incrementally; edits invalidate affected translations and trigger retranslation. To preserve sentence-level quality, inline formatting is stripped before translation, so the current translated view is plain text. The resulting mode is available in every Dossier, translating content lazily as users scroll and updating translated blocks when collaborators edit the document.


### [ClickHouse integrates with Lakehouse Runtime Catalog](https://yomu.fyi/post/clickhouse-integrates-with-lakehouse-runtime-catalog.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Melvyn Peignon
- Published: Apr 22, 2026
- Type: Announcement

ClickHouse 26.2 introduced a beta integration with Google's Lakehouse Runtime Catalog, enabling direct queries on Google Cloud Lakehouse Apache Iceberg tables without data movement or metadata syncing. The system communicates with Google Cloud Storage through the Iceberg REST Catalog endpoint at BigLake using Google Application Default Credentials for OAuth authentication. Users create a database using the DataLakeCatalog engine configured either with a local credentials JSON file or with explicit inline OAuth parameters. Once connected, analysts can query external Iceberg tables directly or ingest datasets into native MergeTree tables for repeated, low-latency workloads. Upcoming enhancements will add write capabilities for Iceberg tables and dedicated connection management tools inside the ClickHouse Cloud interface.


### [Multimodal Data Integration: Production Architectures for Healthcare AI](https://yomu.fyi/post/multimodal-data-integration-production-architectures-for-healthcare-ai.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Maks Khomutskyi
- Published: Apr 22, 2026
- Type: Problem & solution

Production healthcare AI often depends on combining genomics, imaging, clinical notes, and wearables, but separate modality stacks can leave initiatives stalled before deployment. A lakehouse pattern addresses this by landing each modality in governed Delta tables under Unity Catalog controls while supporting cross-modal features, lineage, time travel, and reproducible training sets. Genomics processing can use Glow on Spark, imaging embeddings can feed governed tables and vector search, notes can become temporally aware entities, and wearables can flow through Lakeflow Spark Declarative Pipelines for streaming feature windows. The post compares early, intermediate, late, and attention-based fusion, matching them to dimensionality, temporal, and availability constraints, with late fusion presented as a practical baseline when modalities are missing. It argues that modality masking, sparse attention, and transfer learning are recommended responses because incomplete data is the default and connects the architecture to governed tumor board review, faster cohort assembly, and more clinical workflows.


### [How Yelp Keeps Server-Driven UI Consistent Across Four Platforms](https://yomu.fyi/post/how-yelp-keeps-server-driven-ui-consistent-across-four-platforms.md)
- Company: [Yelp](https://yomu.fyi/company/yelp.md)
- Author: Radu Comaneci, Software Engineer
- Published: Apr 22, 2026
- Type: Problem & solution

Yelp maintains two major applications across Web, iOS, and Android, which previously led to fragmented server-driven UI implementations with mismatched backend component representations. To unify its server-driven UI framework, CHAOS, with its cross-platform design system, Cookbook, Yelp built an automated bridge library called Konbini. Konbini reads centralized JSON interface definitions through Jenkins pipelines to automatically generate synchronized libraries for Python, Kotlin, Swift, and TypeScript. These generated libraries handle component serialization on the backend and deserialization on client platforms. To maintain backward compatibility when component interfaces introduce breaking changes, client requests pass versioned spec identifiers, prompting the backend to execute migration methods that downcast newer component models to older supported versions.


### [Index sharding in ClickHouse Cloud: Petabyte-scale data needs petabyte-scale indexing](https://yomu.fyi/post/index-sharding-in-clickhouse-cloud-petabyte-scale-data-needs-petabyte.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: James Cunningham
- Published: Apr 21, 2026
- Type: Problem & solution

In large-scale ClickHouse Cloud deployments, replicas traditionally load entire primary and secondary indexes from object storage into working memory, creating massive redundant memory overhead as fleets expand. Index sharding resolves this bottleneck by distributing the index analysis phase across available replicas using consistent hashing over data parts. Instead of every node loading full index structures, each replica loads and analyzes only its assigned fraction of primary keys, bloom filters, text indexes, or vector indexes. The query initiator merges the resulting granule match ranges, allowing parallel replicas to stream data without any single machine retaining the whole index. This approach freed working memory for query execution and accelerated index analysis by up to 7.7x in benchmark tests on a 50 billion row table.


### [Real-Time Decisioning for AI Agents: Why you Need a Customer Context Layer First](https://yomu.fyi/post/real-time-decisioning-for-ai-agents-why-you-need-a-customer-context-la.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Alex Dean
- Published: Apr 21, 2026
- Type: Explainer

Real-time decisioning for AI agents depends on more than customer records: it requires a customer context layer that captures what people are doing now and connects that behavior to identity. The post positions the data platform as the center of a composable martech architecture, with agents and applications operating on shared data in open formats such as Apache Iceberg and Delta Lake. Snowplow’s described approach places structured event collection, schema validation, enrichment, and identity resolution before data reaches the platform, while operating in the customer’s cloud environment. It distinguishes historical profiles from granular behavioral event streams and argues that agent interaction outcomes should return to the foundation as first-class events. The resulting four-stage loop—collect, resolve and enrich, serve, and learn—supports simultaneous real-time and historical context, with decision quality dependent on source data coherence.


### [Keeping the Cursor app stable](https://yomu.fyi/post/keeping-the-cursor-app-stable.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Andrew Chan & Kevin Nguyen
- Published: Apr 21, 2026
- Type: Explainer

Cursor's desktop app faces disruptive crashes as usage and features such as subagents, instant grep, and browser use have expanded, with most failures caused by out-of-memory conditions. The team measures instability across Electron's multi-process architecture using telemetry, per-session and per-request crash rates, minidumps, stack traces, crash breadcrumbs, and dashboards that update within minutes. Its two-pronged debugging strategy combines feature-level experiments and oversize-message instrumentation with bottom-up analysis based on a crash watcher, Chrome DevTools Protocol (CDP) stacks, automated daily investigations, opt-in heap snapshots, and low-rate allocation profiling. Findings led to chunking large data, killswitches, lifecycle fixes for leaks, and process isolation for extensions, while prevention uses Bugbot rules, stress tests, garbage collection, performance tests, and automated rollbacks. OOM-per-session fell 80% from its late-February peak and OOM-per-request fell 73% since March 1.


### [AI and the Future of Cybersecurity: Why Openness Matters](https://yomu.fyi/post/ai-and-the-future-of-cybersecurity-why-openness-matters.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Margaret Mitchell, Yacine Jernite, Clem 🤗
- Published: Apr 21, 2026
- Type: Explainer

Mythos is presented as a frontier AI system that rapidly finds and patches software vulnerabilities, with the post arguing that its surrounding system matters more than the language model alone. Its recipe combines substantial compute, software-relevant training data, vulnerability-probing and patching scaffolding, speed, and some autonomy; together, these elements can uncover vulnerabilities, find exploits, and build patches. The post argues that open code and tooling can distribute detection, verification, coordination, and patch propagation across communities, while reducing capability asymmetry between attackers and defenders. For defensive use, it recommends semi-autonomous agents with prespecified actions, human approval, auditable logs, and private deployment using open components and existing security tools. This approach is especially relevant to high-stakes organizations that need inspectable, controllable systems and protection for sensitive data.


### [ClickHouse expands strategic collaboration with Google Cloud, bringing deeper integration, cloud flexibility, and next-generation performance](https://yomu.fyi/post/clickhouse-expands-strategic-collaboration-with-google-cloud-bringing.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Tanya Bragin
- Published: Apr 21, 2026
- Type: Announcement

ClickHouse announced an expanded strategic collaboration with Google Cloud introducing four core architectural milestones. First, native integration with Google Cloud Lakehouse allows direct sub-second queries across data lakes and ClickHouse deployments without ETL pipelines or data duplication. Second, the Bring Your Own Cloud deployment model lets organizations run managed ClickHouse Cloud inside their own Google Cloud Virtual Private Cloud to retain data sovereignty and network controls. Third, ClickHouse Cloud is migrating automatically to Google's custom Arm-based Axion processors, boosting query throughput and cost efficiency for analytical workloads without requiring application changes. Finally, a new ClickHouse MCP server integration connects ClickHouse with Google Antigravity to enhance developer and analyst workflows.


### [From Static Policies to Autonomous Insurance: How AI enables Real-Time Coverage](https://yomu.fyi/post/from-static-policies-to-autonomous-insurance-how-ai-enables-real-time.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Amee Vora, Anindita Mahapatra, Marcela Granados
- Published: Apr 20, 2026
- Type: Explainer

Insurance products often rely on annual pricing, periodic risk assessment, and coverage changes tied to major life events, even though driving behavior, digital fraud exposure, and travel risk can change in seconds. The post presents embedded insurance and usage-based insurance as a shift toward contextual, just-in-time coverage delivered through apps, checkout flows, vehicle purchases, and partner platforms. Its automotive example combines telematics, geospatial data, and real-time weather signals to update risk scores and premiums from trip behavior such as braking, following distance, mileage, location, and environmental conditions. A Databricks Lakehouse architecture registers devices, ingests telemetry through Zerobus, curates data in a governed Medallion architecture, and exposes features through Lakebase and Databricks One, with Unity Catalog supporting access controls, lineage, explainability, and audit trails. The proposed trajectory leads from embedded coverage toward autonomous insurance whose policies continuously adapt to risk.


### [Mercedes-Benz builds a cross-cloud data mesh with Delta Sharing and intelligent replication, cutting costs by 66%](https://yomu.fyi/post/mercedes-benz-builds-a-cross-cloud-data-mesh-with-delta-sharing-and-in.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Alexander Summa, Aleksandar Dragojevic
- Published: Apr 20, 2026
- Type: Problem & solution

Mercedes-Benz built a cross-cloud data mesh to share after-sales data between AWS and Azure while supporting its data-defined vehicle strategy. Existing direct queries incurred egress costs, and weekly full loads delayed updates for datasets that included vehicle over-the-air events and workshop visits; source tables were in Iceberg while many consumers expected Delta. Using Unity Catalog and Delta Sharing, the company federated AWS Glue tables, established cross-metastore shares, and used periodic serverless Sync Jobs with Delta Deep Clone to incrementally replicate selected tables into Azure object storage for local querying. This hybrid model preserves direct sharing for fresher workloads while reducing transfer for less time-sensitive ones. For the initial ten data products, egress costs fell 66%, update frequency improved from weekly to every second day, and the source reports an approximately 93% annual reduction in its 50-use-case comparison.


### [“A generational leap”: How Trio unified payment analytics and cut storage by 88% with ClickHouse Cloud](https://yomu.fyi/post/a-generational-leap-how-trio-unified-payment-analytics-and-cut-storage.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Apr 17, 2026
- Type: Problem & solution

Brazilian payment processor Trio faced severe data challenges after scaling and acquiring PayBrokers, processing over 243 million payments in early 2025 across disparate legacy databases. Because financial reconciliation and compliance demand absolute precision, the engineering team unified their analytics layer by migrating to ClickHouse Cloud. To handle out-of-order and duplicate events from their Redpanda streaming pipelines, Trio implemented sliding-window refreshable materialized views alongside ReplacingMergeTree engines rather than recomputing entire historical datasets. A custom ETL service manages schema drift before inserting clean records into ClickHouse, while batch ingestion is coordinated through Airflow. Consequently, Trio reduced storage consumption by approximately 88% through columnar compression, ingested a 5-billion-row migration without issue, and maintained real-time query performance across more than one billion daily ledger data points.


### [How we built Stripe Credits: A programmable, auditable way to pay your Stripe fees](https://yomu.fyi/post/how-we-built-stripe-credits-a-programmable-auditable-way-to-pay-your-s.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Pratik Gupta
- Published: Apr 16, 2026
- Type: Explainer

Stripe Fee Credits replaces ad-hoc discounts and manual cash adjustments with a programmable, auditable virtual balance dedicated to eligible Stripe fees. The architecture has three layers: a standalone gRPC credits engine for metadata, policies, lifecycle, and audit history; logical financial accounts built on Money Movement and Storage; and fee-settlement orchestration that applies eligible credits before the primary balance. Logical accounts carry ledger balances without representing real cash, while platform checks restrict Originated Money Transfers to logical-to-logical movements and exclude these accounts from cash sweeps. Prepaid and Stripe-funded credits share the user experience but use separate funds flows: prepayments create deferred revenue and collect tax at drawdown, whereas Stripe-funded credits are booked as Stripe costs with taxes written off. The resulting infrastructure supports incentives, prepaid annual fees, and other programs across 26+ countries, with Dashboard balances and real-time auditability, while avoiding bespoke engineering for each commercial model.


[Newer posts](https://yomu.fyi/topic/architecture/page/20.md) · [Older posts](https://yomu.fyi/topic/architecture/page/22.md)