Loading…
Browse by type
Problem & solution reads
Problem & solution posts on Yomu, summarised newest first.
- 916 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.
Problem & solution posts
Building Production-Scale Visual AI for Claims and Returns at Swiggy Instamart
Swiggy Instamart built two vision-language-model systems to handle more than 40,000 daily claims involving wrong, expired, poor-quality, damaged, and packaging-related items, while reducing reliance on human verification. The Image Validation Model (IVM) compares customer uploads with catalog images, routes each category through a specialised LangGraph StateGraph handler, and combines its structured verdict with fraud, confidence, and business-rule signals. Reverse Quality Check (RQC) compares delivery executives’ pickup photos with the original claim images, allowing matching returns to proceed and prompting retries when they do not. The architecture keeps visual assessment probabilistic but makes routing, aggregation, escalation, and financial decisions deterministic, with traces for category, outputs, confidence, cost, and latency. After more than six months in production, the systems are associated with a 6% customer-satisfaction increase, 20,000 fewer weekly Image Validation tickets, 98.5% RQC accuracy, and over 5,000 daily RQC tickets resolved without human intervention.
How Wayfair cut ML model costs by 90% (twice!) with Cursor
Wayfair’s Applied Research team used Cursor to automate and parallelize experiments for a product-tag validation model that checked attributes against images, descriptions, and reviews. The model was accurate but too expensive to run across the catalog. In December 2025, five researchers standardized the dataset and evaluation benchmark, then built and tested 110 distinct variants in four days; the winning architecture reduced inference costs by 94%, improved precision, and became the production baseline. In March 2026, the team benchmarked against that baseline, ran 140+ new experiments, and added genetic-algorithm searches, achieving another 90% cost reduction. The workflow used 20+ parallel agents, desktop and CLI surfaces, and cloud agents, leaving researchers to formulate hypotheses, interpret results, and choose candidates while Cursor implemented, ran, and measured variants.
Teaching Sidekick to say no: automated data curation with LLM judge consensus
Sidekick’s customer-segmentation skill was fine-tuned on tens of thousands of successful, de-identified production queries, so it generated zero-result queries instead of refusing impossible requests such as finding customers by occupation. A small Toloka dataset of roughly 600 standard queries and 602 refusal annotations became a seed for four frontier LLM judges, which were calibrated against human-labeled examples and required to agree on both verdict and reasoning. The pipeline used mutually exclusive categories to resolve conflicting labels, filtered disagreements for human review, and fed accepted examples from later production traffic back into fine-tuning. With automated curation, the segmentation pass rate rose from 0.762 to 0.798 versus naive merging, while manual validation measured 86.3% refusal accuracy and a 4.6% false-positive rate. The resulting flywheel is being applied to other Sidekick skill models and data-quality problems.
Cisco Talos empowers threat researchers while reducing TCO by 75% with ClickHouse Cloud
Cisco Talos operates a threat-intelligence reputation service that classifies file hashes from Cisco product telemetry, enabling researchers to make security decisions about benign and potentially malicious artifacts. Previously, a custom pipeline moved data from Databricks Delta and S3 through EKS, where staging and aggregated tables supported an API; the self-managed design retained about a petabyte of compressed, replicated data on 32 high-compute nodes. The team rebuilt the system in four phases, using Terraform, SSO/SAML, sharded ClickPipes for a 12-month backfill, shadow traffic, and a validated cutover to ClickHouse Cloud on AWS. The resulting architecture uses EventBridge, SQS, ClickPipes, SharedMergeTree, object storage, and AWS PrivateLink, delivering less than a minute of queue lag, zero data loss, a 100% response match across more than 1,000 hashes, zero downtime, roughly 90% lower storage costs, and about 75% lower TCO while reclaiming over 300 engineer hours annually.
Better, faster, less wrong: Enhancing issue grouping
Sentry upgraded its AI-driven issue grouping system to better prevent duplicate issues without merging distinct application errors. The team trained lightonai/modernbert-embed-large using Matryoshka Representation Learning on hundreds of thousands of stacktrace pairs labeled by Claude Sonnet 4.5. To optimize the high-throughput ingestion pipeline, embeddings were truncated from 768 to 64 dimensions, combined with bfloat16 precision, PyTorch SDPA, and CUDA graph compilation. A phased live rollout used threshold-gated index backfilling and automatic fallback to prevent issue spikes during migration. In production, the v2 model cuts the overgrouping rate from 8% to 4%, increases prevented duplicate issues to 70%, and delivers 6x faster inference.
How ERGO Hestia reduced time-to-market with Databricks Lakebase and Model Serving
ERGO Hestia redesigned its real-time pricing platform to reduce deployment friction across more than 100 models and 1,000 variables while preparing B2C capabilities. Previously, processed data moved from Databricks through extraction jobs, external Azure PostgreSQL, and a custom caching adapter, creating governance overhead, deployment coordination, and latency spikes during large refreshes. The new architecture uses Lakebase Sync Tables as an online serving layer and Databricks Model Serving Endpoints, keeping data, request logic, and model serving within the lakehouse; Unity Catalog supplies lineage, version tracking, access controls, and audit trails. An incremental migration started with a low-criticality endpoint, measuring 20ms latency and less than 5% CPU utilization at 40 requests per second, before expanding toward larger workloads and the planned decommissioning of PostgreSQL.
How Ecolab rebuilt retail intelligence on Databricks and Anthropic Claude
Ecolab needed to combine audits, health inspections, pest telemetry, and other data from nine systems so retail teams could answer location-specific compliance questions. Its Retail Intelligence application is a native Databricks App using Lakebase Postgres, Lakeflow, and Spark Declarative Pipelines to move governed data into a Unity Catalog lakehouse, while Foundation Model APIs serve Claude Sonnet, Claude Haiku, and Gemini. A Coordinator Agent delegates requests to specialized agents that use Vector Search, SQL, Unity Catalog Functions, and an external MCP server; a Response Agent returns cited answers, with short- and long-term memory stored through Lakebase. The system also applies five Judge LLMs, MLflow tracing, and ai_query() batch inference. Report preparation fell from two weeks to under two minutes, while the assistant supports approximately twelve languages at about 98% accuracy.
How to make the case for giving your AI Agent system access
AI Agents can answer support questions without backend access, but they cannot complete changes such as payment-plan updates, claims, or account edits. Connecting an Agent to CRM, billing, order-management, or other systems moves support from explanation to end-to-end resolution, and the source frames that shift as the basis for an engineering request. Data from a 2026 Customer Service Transformation Report shows improved metrics among 87% of teams with mature AI deployment, versus 62% overall, while only 10% report reaching that stage; an internal comparison found Procedures with system access produced the largest gains in workflows requiring judgment, branching, live data, or handoffs. The recommended case starts with one high-volume, repeatable workflow, maps its data and actions, and defines a narrow pilot. Teams can then progress from no integration to read-only access and finally write actions, using measured results to justify further work.
Metric Semantic Layer: How Lyft Governs and Scales Key Data Definitions
As Lyft scaled, different teams developed conflicting definitions for key business metrics due to the lack of centralized version control and shared standards. To resolve this, Lyft implemented an internal Metric Semantic Layer as a Python package that stores authoritative metric definitions in YAML files with Jinja SQL templates. The system restricts onboarding to Golden Metrics used across multiple applications, requiring team-based approval from both Business and Operational Owners for any definition changes. Standardized definitions are exposed through Python APIs, integrated into the Amundsen data catalog and self-service user interfaces, and surfaced to AI tools via a Model Context Protocol.
AI Serving Platform That Adapts to Your Model
Databricks Custom Model Serving addresses the operational burden of serving custom models, whose resource profiles, traffic patterns, and latency requirements vary widely from small CPU classifiers to large GPU-backed language models. Its fully managed platform packages MLflow models and uses isolated Kubernetes deployments, model-appropriate runtimes, and a short request path to limit interference and per-request overhead. At the center, the AutoPilot Pod Autoscaler combines active concurrency and queue signals for horizontal scaling with CPU, GPU, and memory measurements for model-aware target-concurrency adjustment, allowing one controller to adapt across workloads. Warm pools, provisioned concurrency, and zero-downtime updates address startup and deployment concerns, while reported production results include 90%+ cost savings for some customers, up to 2x improvement in p99 and p50 latency, 100K+ QPS, and 99.99% availability.
Encoding Your Domain Expert: The Context Layer Behind Spotify's Data Assistant
Spotify developed an AI data assistant, Vedder, to scale access to over 70,000 internal datasets without overwhelming domain experts. Because raw warehouse schemas omit critical semantic nuances and overwhelm LLM context windows, engineers implemented a domain-owned context layer organized into clusters. Each cluster contains profiled datasets, expert-curated business documentation, and vetted question-and-SQL pairs that guide a ReAct-based agent. Automated query history extraction proved insufficient on its own, as domain curators accepted only 12.5% of inferred query pairs due to noise. Deployed across Slack, IDEs, and a dedicated web interface, the system maintains reliability through continuous cluster health monitoring that flags schema drift and degraded examples.
Modern BSA/AML compliance on Databricks
AML operations are strained by fragmented systems, high false-positive volumes, manual case documentation, and opaque vendor scoring, leaving analysts focused on backlog rather than financial-crime intelligence. The proposed Databricks Data + AI Platform unifies transaction monitoring, KYC, sanctions, case history, and policy data under Unity Catalog, using Lakeflow Connect and a Bronze–Silver–Gold Delta architecture with masking, row-level security, and lineage. MLflow, Model Serving, Lakehouse Monitoring, and inference tables support institution-specific detection models, while Agent Bricks coordinates agents for evidence gathering, recommendations, and SAR drafting with analysts retaining final decisions. The architecture also uses Lakebase for governed operational state and Databricks Apps for analyst and executive experiences. Reported outcomes include a 75% reduction in false positives reaching the analyst queue and compressing three-to-six-hour investigations to minutes.
From Chaos to Clarity: How We Built a Unified, Self-Routing Support Ops Ticketing System at Lyft
Lyft Urban Solutions' Support Ops team faced fragmented ticketing across four Jira Help Centers with duplicate intake forms, lack of routing logic, and no dashboards. Over five years, the team redesigned the intake architecture by replacing redundant forms with dynamic Jira Proforma forms featuring conditional branching. Automated Jira rules were introduced to assign tickets, apply a taxonomy of over 30 labels, and trigger PagerDuty alerts for P0 incidents. To unify operations across disparate backend projects, the system automatically clones and links tickets from a single portal to downstream project boards. Jira Structures and a Jira-to-Mode ETL pipeline provided cross-project rollups and business analytics before an upcoming Jira Cloud migration required dashboard rebuilds.
amazon.science
Real-world grounding in agentic AI
Foundation models deployed as physical AI agents face critical risks when natural language hallucinations translate into physical law violations and operational hazards. To ensure physical consistency and reliability, researchers from the University of California, San Diego and Amazon Fulfillment Technology introduced four grounding pillars. Physics-guided deep learning embeds symmetries and differential equations into pretraining to obey conservation laws while reducing dataset requirements. The UQ4CT framework applies mixture-of-experts architectures to produce calibrated uncertainty estimates, cutting expected calibration error by over 25% and halting operations when safety thresholds are breached. Furthermore, the Adapting-While-Learning framework and formal verifiers like Hilbert integrate numerical simulators and interactive proof systems to enforce mathematical and scientific correctness.
Transforming solar and wind maintenance reports with Genie and AI agents
Plenitude and Databricks built an agent-based system that turns solar and wind plant maintenance PDFs into structured data for cross-plant analysis. Event-driven ingestion uses Databricks Jobs and the ai_parse_document AI Function to extract text, tables, figures, and metadata, then stores page- and object-level JSON records in Delta Lake with coordinates, version history, and links to source reports. A Genie space uses Unity Catalog metadata, knowledge-store instructions, and SQL generation to answer natural-language questions, produce visualizations, and export results, while Agent Bricks can orchestrate multi-step workflows and downstream actions. The design also applies automatic liquid clustering to dynamic queries and row-level security to restrict results by country. The resulting data layer supports historical trends, plant comparisons, recurring-fault analysis, and a foundation for predictive maintenance, although the source frames predictive use as a future improvement.
amazon.science
Bridging intent and execution in agentic systems
AI agent performance increasingly depends on the harness mediating interactions between language models and tools rather than raw model reasoning. To address mismatches between model intent and harness execution, the authors developed Simple Strands Agent, an open-source single-agent harness. The architecture implements robust tool-interface practices, such as condensing middle log outputs from bash commands, enforcing line-aware string replacements with ambiguity checks, and returning diff files after edits for verification. Evaluated across benchmarks like SWE-Bench-Verified, SWE-Bench-Pro, and Terminal-Bench 2, these model-harness alignments yielded consistent performance gains across frontier models without task-specific tuning. The findings show that model families require tailored interface codesign alongside stable evaluation infrastructure to achieve optimal benchmark reliability.
Works on my machine: how we use AI to reproduce reported bugs
Sentry maintains 159 open-source SDK packages across multiple languages, leading to high triage volume and significant boilerplate setup overhead when reproducing reported bugs. To streamline this workflow, the team developed a Claude skill that parses GitHub issue URLs to extract language, framework, and SDK versions. The skill generates a reproduction branch, sets up minimal boilerplate using ecosystem package managers like uv or npm, attempts execution, and opens a pull request with explicit run instructions. By allowing the model to exit early when encountering overly complex environments, the tool reliably prepares reproducible playgrounds without stumbling on edge cases. This workflow saved several hours in testing scenarios like Django HTTPS proxy bugs while avoiding excessive bot noise through scoped human interaction.
Building Agents that Don't Break Themselves
Agents that call shell commands can damage their host, so the post separates the agent’s durable home from the environment where untrusted code executes. It presents Sprites as disposable, isolated sandboxes, with one Sprite per session for the multi-user SpriteDoc and one persistent Sprite per task for Hermes Agent. SpriteDoc injects each user’s flyctl credential only for the command that needs it, avoiding token storage while preserving the user’s permissions; Hermes can omit approval prompts because the sandbox protects the host. The same separation applies when Hermes itself runs inside a Sprite: its commands are dispatched to another Sprite with a different identity and boot. Copy-on-write checkpoints provide an undo path, restoring deleted files and tools in about nine seconds.
Building a File Semantic Analyzer: Guarding Outbound Data at Scale with AI
Organizations struggled to determine whether files leaving their environment contained business-critical information, personal data, or non-essential content. Uber’s File Semantic Analyzer (FSA) addresses the limits of keyword- and pattern-based DLP by extracting text from documents, using OCR for image-based files, chunking large inputs, and sending context-rich text to a fine-tuned large language model. The model produces summaries, named entities, key phrases, inferred intent, classifications, and explanations, which a rule-based engine uses for policy enforcement while analysts validate high-fidelity findings and provide feedback. The system reduced incident-response time from multiple hours to a few minutes per case, saves an average of five minutes per file, and is expected to analyze 150,000 files annually; the post reports a 97% reduction in false positives.
Agentic Supplier Management with MongoDB Atlas, Voyage AI, and Multi-Modal Search
Retailers face major operational bottlenecks when managing supply chain disruptions through legacy enterprise resource planning systems and siloed documents. Traditional tabular structures struggle to index multi-modal data like images and PDF contracts, delaying critical supplier discovery during regional crises or tariff changes. To overcome these constraints, organizations decouple supplier management into a dedicated operational data layer using MongoDB Atlas and Voyage AI. This architecture stores polymorphic supplier profiles and high-dimensional vector embeddings within a unified collection, enabling natural language semantic search for alternate partners. Real-time updates propagate across systems using MongoDB Change Streams, allowing businesses to rapidly assess supply chain impacts and identify alternative vendors in minutes.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.