Offshore EPC Tender Intelligence System with LlamaIndex
Qatar's massive offshore oil field expansion tenders feature thousands of technical specifications, HSE addendums, and commercial terms. Build an advanced RAG and agentic evaluation framework using LlamaIndex with hybrid vector/BM25 retrieval. Achieve a retrieval Mean Reciprocal Rank (MRR) above 0.85 on engineering query evaluations.
What you are building
The core problem, expected build, and operating context for this challenge.
Construct an agentic RAG index using LlamaIndex with hybrid retrieval and reranking to answer technical queries on Qatar offshore EPC tender documents.
How work is evaluated
Measures retrieval MRR and answer correctness on complex offshore EPC tender queries.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
source_citation_presence
Checks that synthesized response includes explicit document page citations
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
retrieval_mrr
Mean Reciprocal Rank of relevant chunks in top-5 retrieval results • target: 0.85 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement hierarchical document indexing and hybrid vector + BM25 search in LlamaIndex
Incorporate reranking models (e.g., Cohere or sentence-transformers) for context optimization
Build dynamic Query Engine Tools integrated into a LlamaIndex agent workflow
Evaluate retrieval quality using TrueLens or LlamaIndex RAG evaluation metrics
Reference links and supporting material
Dataset of 50 synthetic offshore engineering specification documents covering jackets, topsides, and subsea pipelines.
Challenge setup
Measures retrieval MRR and answer correctness on complex offshore EPC tender queries.
Challenge input
JSON object with tender question string and target document collection ID
LlamaIndex
Leading framework for structured data indexing and retrieval synthesis.
Evaluated output
JSON containing synthesized answer, cited source document IDs, and confidence score
- Checks that synthesized response includes explicit document page citations
- Mean Reciprocal Rank of relevant chunks in top-5 retrieval results • target: 0.85 • range: 0-1
- Benchmark: RAGBench-Industrial
- Retrieval Mrr target: 0.85
- 1 public reference case
- 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
- Llama Index · Required
- Traceloop · Optional
- Deepgram · Optional
Evaluation contract
- source_citation_presence · Weight 1
- retrieval_mrr · Weight 1
Metrics
- retrieval_mrr · Target 0.85
Benchmarks
- RAGBench-Industrial
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with tender question string and target document collection ID
Output: JSON containing synthesized answer, cited source document IDs, and confidence score
Execution harness: Python
Evaluation tools
- LlamaIndex · RAG Framework
Recipe freeze marker
No recipe freeze time is recorded.