Overview
Pilot deployment with real Walmart logs and statistical tests gives moderate-to-strong evidence. Internal data and closed code limit external reproduction.
Citations0
Evidence Strength0.80
Confidence0.80
Risk Signals9
Trust Signals
Findings with numeric evidence: 4/4
Findings with evidence refs: 4/4
Results with explicit delta: 7/7
Reproducibility
Status: No open assets linked
Open source: Unknown
At A Glance
Cost impact: 60%
Production readiness: 70%
Novelty: 45%
Why It Matters For Business
You can keep advanced agentic reasoning for hard requests while giving fast answers for common lookups. That reduces user wait time and session rounds, which likely raises engagement and lowers operational cost.
Who Should Care
Summary TLDR
AdaptJobRec is an agentic conversational job recommender built for Walmart. It classifies incoming queries as simple or complex. Simple queries bypass planner/memory and call fast APIs; complex queries use a few-shot memory filter and a nested task planner that groups parallel subtasks. On Walmart data, this routing plus personalization cuts response latency by ~53% and reduces dialogue rounds while slightly improving ranking metrics.
Problem Statement
Agentic conversational recommenders give richer answers but are slow. Simple queries (e.g., 'check application status') waste time if they always trigger planner and memory modules. The paper asks: can we keep agentic reasoning for hard queries while giving fast responses for simple ones?
Main Contribution
AdaptJobRec: an LLM-powered agent that classifies query complexity and routes simple queries directly to fast tools while reserving full agentic flow for complex queries.
Few-shot memory processing module that filters chat history for only relevant segments, reducing redundant planning.
Key Findings
AdaptJobRec cuts average response latency by about half compared to a RAG baseline in pilot users.
AdaptJobRec reduces the number of conversation rounds needed to get the target information.
Results
| Metric | Value | Baseline | Delta | Split / Dataset | Evidence | Evidence Ref |
|---|---|---|---|---|---|---|
| Hit@10 (job recommendation) | 0.3176 | Plan & Execute 0.3127 | +0.0049 | 10,014 Walmart users | Table 1: AdaptJobRec vs baselines | Table 1 |
| NDCG@10 (job recommendation) | 0.0810 | Plan & Execute 0.0799 | +0.0011 | 10,014 Walmart users | Table 1: AdaptJobRec vs baselines | Table 1 |
What To Try In 7 Days
Add a lightweight complexity classifier to route simple queries straight to existing APIs.
Implement a small memory filter (few-shot prompt) that returns only relevant chat snippets for complex queries.
Cache frequent tool responses (Redis) to avoid repeated LLM calls for high-frequency lookups.
Agent Features
Memory
Planning
Tool Use
Frameworks
Is Agentic
Yes
Architectures
Optimization Features
Token Efficiency
Infra Optimization
System Optimization
Inference Optimization
Reproducibility
Risks & Boundaries
Limitations
Evaluation uses internal Walmart data and a small pilot (150 sessions); external generalization is untested.
System depends on a rich People.AI knowledge graph — missing or sparse KG coverage will hurt recommendations.
When Not To Use
You lack a structured knowledge graph or matching recommendation APIs.
Most queries are complex and require deep planning in every session (less benefit from routing).
Failure Modes
Misclassifying a complex query as simple leads to incomplete answers.
Memory filter omits relevant history, causing planner to miss necessary context.