Submission type
Coded Agent with UiPath SDK
Name
Kalyan Gande
Industry category in which use case would best fit in (Select up to 2 industries)
Logistic
Manufacturing
Marketing/Sales
Complexity level
Advanced
Summary (abstract)
The Inventory Forecast Anomaly Guard is a unified LLM-driven agent built using UiPath’s LangChain SDK that detects and acts on inventory imbalances in near real-time.
It automatically downloads structured stock data from UiPath Buckets, computes deviation and stockout metrics, performs single-pass LLM reasoning to classify anomalies, and triggers reorder or escalation actions through UiPath Orchestrator and Action Center — enabling early detection, traceable AI-based recommendations, and reduced human effort in supply chain monitoring
inventory-anomaly-guard-plan
Detailed problem statement
Organizations face frequent stockouts and overstock situations due to poor demand sensing and delayed response to anomalies in the inventory pipeline.
Traditional dashboards highlight deviations but lack contextual reasoning or automated interventions. Manual monitoring across thousands of SKUs leads to:
Delayed reaction to stockouts
Overstock causing high carrying cost
Lack of visibility into why an anomaly occurred
Inconsistent manual escalation workflows
The challenge was to build an agentic system that could analyze, reason, and act within a single execution cycle — providing both insights and automated actions for critical inventory anomalies
inventory-anomaly-guard-plan
Detailed solution
The solution implements a UiPath LangChain agent orchestrating a stepwise flow:
Download File – Fetch latest SKU inventory data from UiPath Buckets.
Parse & Normalize – Convert CSV/XLSX rows into structured SKU metrics.
Compute Metrics – Derive deviation %, days-to-stockout, and over-safety ratios.
LLM Classification – Invoke GPT-4o-mini or GPT-4.1 model for single-pass structured output classification (none, watch, warning, critical).
Action Planning – Automatically decide between reorder process or Action Center escalation.
Execution Loop – Trigger UiPath process or create escalation task until queues are cleared.
Finalize Output – Return summarized JSON with anomalies, actions, metrics, and LLM rationale
Architecture Flow:
download_file → parse_rows → compute_metrics → llm_classify_and_plan → execute_actions → finalize
LLM Rationale Traceability:
Every decision includes llm_confidence, reason, and a global llm_rationale for auditability.
Observability:
The agent emits telemetry (token counts, severity distribution, action stats) enabling performance analytics via UiPath Insights
inventory-anomaly-guard-plan
Narrated video link (sample: https://bit.ly/4pvuNEL)
Expected impact of this automation
The solution significantly improves operational efficiency by automating the entire anomaly-triage and response workflow. This reduces the need for manual reviews by roughly 60–70%, freeing teams from repetitive validation tasks.
It enhances business continuity by identifying potential stockout risks early in the process. The system is able to detect and cover nearly 80% of all actual stockout events, ensuring proactive mitigation.
From a cost-optimization perspective, the solution helps prevent excess inventory and unnecessary carrying costs. By enabling smarter forecasting and anomaly detection, organizations can achieve 15–20% savings in overall inventory-related expenses.
The system also strengthens auditability and explainability. Each SKU-level decision is accompanied by a logged LLM-generated rationale, providing 100% traceability for every anomaly detected or action taken.
Finally, the solution is designed for agentic readiness, integrating seamlessly with UiPath Orchestrator. This creates a single, unified execution loop that eliminates manual handoffs entirely, enabling a fully automated, end-to-end workflow.
UiPath products used (select up to 4 items)
UiPath Action Center
UiPath Apps
UiPath Coded Agents
UiPath Orchestrator
Integration with external technologies
LangGraph / LangChain SDK: Orchestrates the agent’s flow using a structured state graph. OpenAI / Azure OpenAI Models: Provides LLM reasoning with strict JSON-schema outputs. UiPath Buckets: Pulls inventory files directly from cloud storage. Action Center APIs: Sends escalations to humans through interrupt-based actions. OpenInference / OTEL: Captures telemetry for tracing tokens, anomalies, and execution time. Python Pandas / Pydantic: Cleans, normalizes, and validates CSV/XLSX data with strict typing.
Hi there, 
Nov 5–16: Jury evaluation by 
Nov 17: Winners announced 
.
,