Dear Algo: A Precision-First Agentic Intent Layer for Unified Search and Recommendation

2026-08-16

AI Summary

This study addresses the challenge of unified information discovery arising from divergent intent encoding in search and recommendation systems. We propose an Agent Intent Layer that compiles multidimensional intents into executable plans, establishing a natural language-driven, precision-first unified framework via an "Intent-Retrieval Contract." This approach innovatively integrates intent compilation, semantic multimodal reranking, and an LLM-as-a-Judge evaluation mechanism to effectively bridge the semantic gap between heterogeneous systems. Experimental results demonstrate that the proposed method achieves a precise relevance rate of 94.4%, improves candidate quality by 1.11, and reduces the proportion of irrelevant content by 1.97 percentage points. These findings indicate significant enhancements in both information feed relevance and overall user experience, validating the efficacy of aligning disparate system intents through structured compilation and rigorous evaluation.

Abstract

Search and recommendation serve a shared discovery objective but encode intent differently. We study this boundary through Dear Algo on Threads, a deployed product where open-ended requests such as \emph{more NBA news} or \emph{less politics} steer subsequent feed recommendations rather than return a one-shot result list. Its agentic intent layer compiles explicit, inferred, negative, and compound intent into a grounded executable plan, then invokes conventional retrieval and optional semantic or multimodal reranking. The layer shares an intent-to-retrieval contract without requiring one model or serving path across search-like and recommendation-like modes. We evaluate Dear Algo under a precision-first objective. In a blinded audit of 300 public request-item pairs (296 evaluable), a strict categorical LLM-as-a-judge gate achieved 94.4\% exact-Relevant precision [88.8\%, 98.9\%]. Across 72 normalized request clusters, the full configuration produced 7.73 judge-qualified candidates per 20 slots versus 6.61 for an LLM-derived-query baseline, a gain of 1.11 [0.12, 2.12]. In a candidate-randomized serving-path study restricted to the reranker path's first 72 eligible hours, the user-weighted judge-Irrelevant share among judged admissions was 2.80\% versus 4.78\% off (-1.97 points [-3.02, -0.94]), while Exact-Relevant share was 2.24 points higher [0.08, 4.41]. Together, these studies show how explicit natural-language intent can be carried into feed recommendation under a precision-first evaluation framework

Problem

• Search and Recommendation
• Natural Language Intent
• Feed Recommendation
• Precision-First Evaluation

Innovation

• Agentic Intent Layer
• Unified Search and Recommendation
• Precision-First Evaluation
• Intent Compilation
• LLM-as-a-Judge