Skip to main content
Back to challenges
Challenge

Deploy Policy Assistant ALLIE on Modal using LlamaIndex

Carriers need performant RAG engines to deliver policy answer engines safely. You will construct a domain-specific policy QA workflow using LlamaIndex and deploy the inference microservices onto Modal's serverless GPUs. The microservice must achieve under 400ms end-to-end latency while scoring >0.85 on RAG evaluation metrics.

Machine LearningHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build and deploy an enterprise-grade policy assistant microservice using LlamaIndex and Modal for serverless vector retrieval.

Delivery guide

How work is evaluated

Evaluation

Evaluates RAG generation precision, context relevance, and response latency on policy queries.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

latency_under_400ms

Validates that latency_ms is under 400 milliseconds.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

context_precision

Precision of retrieved policy passages • target: 0.85 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Build modular RAG index pipelines using LlamaIndex vector store abstractions.

  • Deploy serverless GPU inference functions on Modal with container caching.

  • Evaluate retriever relevance and hallucination scores using TruLens or Ragas.

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of 200 standard commercial property policy filings, endorsements, and synthetic customer inquiry pairs.

Challenge setup

Evaluates RAG generation precision, context relevance, and response latency on policy queries.

Challenge input

JSON object with query, policy_id, and context_documents

LlamaIndex

Assigned framework for data indexing and retrieval.

Modal

Assigned infrastructure tool for serverless GPU execution.

Evaluated output

JSON object with answer, retrieved_node_ids, and latency_ms

Evaluation checks
  • Validates that latency_ms is under 400 milliseconds.
  • Precision of retrieved policy passages • target: 0.85 • range: 0-1
Configured evaluation evidence
  • Benchmark: Enterprise RAG Speed & Accuracy Benchmark
  • Context Precision target: 0.85
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

Environment
  • Modal · Required
Action Space
  • Llama Index · Optional
  • Exa · Optional
Policy Serving
  • Modal · Required

Evaluation contract

  • latency_under_400ms · Weight 1
  • context_precision · Weight 1

Metrics

  • context_precision · Target 0.85

Benchmarks

  • Enterprise RAG Speed & Accuracy Benchmark
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON object with query, policy_id, and context_documents

    Output: JSON object with answer, retrieved_node_ids, and latency_ms

    Execution harness: Python

Evaluation tools

  • LlamaIndex · RAG Framework
  • Modal · Serverless Runtime

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Deploy Policy Assistant ALLIE on Modal using LlamaIndex