Instacart Feature Trails

Feature trailActive

Agentic Machine Learning Modeling

Instacart is exploring agent-assisted machine learning modeling loops to accelerate the development and iteration of ML models. This approach involves human MLEs defining problems and supervising agents, which are tasked with generating hypotheses, coding solutions, and evaluating them. The system leverages Instacart's in-house ML platform, Griffin, for model estimation and evaluation, and is developing new infrastructure specifically for agentic modeling use cases. Case studies in delivery time and catalog attribute extraction demonstrate improvements in model performance (e.g., 3-5% offline MAE reduction for delivery time, significant recall gains for attribute extraction) and reduced iteration time. The process involves adaptive hypothesis generation, model family exploration, hyperparameter tuning, and combining smaller improvements, mimicking human MLE workflows.

1 post · 2026

Timeline

Feature trailActive

On-call Reasoning Harness

Blueberry is a Slack-native on-call reasoning harness designed to shorten time to first insight (TTFI) and time to test theories (TTTT) for on-call engineers. Blueberry automatically triages alerts within Slack threads, providing grounded explanations by correlating historical data (postmortems, runbooks) with live operational signals and changes. It features a durable, tool-aware architecture with a layered Model Context Protocol (MCP) for shared and team-specific investigators, enabling faster

1 post · 2026

Timeline

Feature trailActive

Experimentation Variance Reduction

Instacart is enhancing its experimentation capabilities by developing novel methods for variance reduction. This includes building predictions at a finer grain (order-level) and aggregating them up, significantly improving statistical power and reducing experimentation time. This approach leverages order-specific covariates to create more accurate pre-treatment predictions, overcoming limitations of traditional region-day CUPED. The system also incorporates a placebo test to ensure feature contamination does not introduce bias. This has led to substantial reductions in experimental runtimes, allowing for more experiments per quarter and faster identification of marketplace improvements.

2 posts · 2026

Timeline

Feature trailActive

High-Cardinality Marketplace Modeling

Instacart is enhancing its experimentation capabilities by developing novel methods for variance reduction. This includes building predictions at a finer grain (order-level) and aggregating them up, significantly improving statistical power and reducing experimentation time. This approach leverages order-specific covariates to create more accurate pre-treatment predictions, overcoming limitations of traditional region-day CUPED. The system also incorporates a placebo test to ensure feature containment. This post details the application of PyFixest to efficiently handle high-cardinality categorical variables in marketplace modeling, enabling more precise analysis of marketplace algorithms and their impact.

1 post · 2026

Timeline

Feature trailActive

Semantic IDs for Product Understanding

Instacart has rebuilt its ads retrieval system by moving from a scoring-based approach to a generative model that spells out product recommendations token by token. This new system leverages Instacart Semantic IDs (SIDs), a compressed product representation generated by an RQ-VAE, which allows for better coverage of the entire catalog, including new products, and enables the model to generalize based on semantic codewords rather than just memorizing specific product co-occurrences. The training process uses a contrastive loss regularized by catalog structure and hierarchical batch sampling. The system also incorporates a two-flavor strategy for precision (ESCI) and discovery (ESCI+Gemma) recommendations.

2 posts · 2026

Timeline

Feature trailActive

AI Impact on Applied Science Roles

Instacart is analyzing the impact of AI on applied science roles, observing a significant increase in overall productivity (PRs and LoC) driven by AI-powered tools. This has led to a redistribution of tasks, with a decrease in the share of work on standardized ML and data pipeline tasks, and an increase in new task categories like Platform/Tooling and Frontend/UI development. This shift allows applied scientists to take on a more diverse set of responsibilities, leveraging AI to overcome previous skill barriers and reduce coordination costs.

1 post · 2026

Timeline

Feature trailActive

Multi-Tenant Marketing Platform

Instacart has developed a multi-tenant marketing platform to enable personalized marketing capabilities for its white-label Storefront Pro retailers. This platform extends existing marketing automation infrastructure by introducing per-retailer workspaces within a third-party vendor, ensuring data isolation and brand integrity. Key components include Instacart tooling for retailers, a Campaigns Engine for audience evaluation and personalization, a stream consumer and batching layer to improve throughput and API efficiency, and a CRM Service that abstracts provider-specific logic. Operational considerations include domain configuration for sender addresses and automated IP warming for new retailers.

1 post · 2026

Timeline

Feature trailActive

Domain Adaptive Learning for Ads

Instacart is enhancing its advertising platform (Carrot Ads) by developing and applying Domain Adaptive Learning techniques. This approach transfers knowledge from the data-rich Instacart Marketplace (source domain) to new partner websites (target domains) to address the 'cold start' problem. This enables performant ad serving even with limited historical interaction data on partner sites. The method involves adapting both the neural network architecture (reusing and fine-tuning embedding layers) and the training data (matching features across domains). The goal is to onboard new retail partners faster and more scalably, delivering performant ad systems from day one.

1 post · 2026

Timeline

Feature trailActive

LLM-Powered Discovery Recommendations

Instacart is transforming its discovery recommendations by leveraging Large Language Models (LLMs). This initiative moves away from static content libraries and human-driven generation towards an AI-native platform for content generation, evaluation, and retrieval. The system employs a top-down, cascaded approach, starting with page design and theme generation personalized to users, followed by retrieval keyword generation, quality and diversity filtering, and finally, product and pagewise ranking.

1 post · 2026

Timeline

Feature trailActive

Caper Data Flywheel and Observability

Instacart built Capsight, an end-to-end platform for Caper smart carts. It acts as a distributed data collection and model improvement engine. The system comprises the Capsight Collector (on-device intelligent data capture), Capsight Depot (cloud platform for data management, ingestion, curation, and AI-assisted annotation), and Capsight Learner (distributed training platform). This enables a continuous feedback loop from edge-captured data to ML training, significantly reducing model iteration time and annotation costs through VLM pre-labeling. The system is expanding to full sensor fusion for a foundation model understanding store environments.

1 post · 2026

Timeline