Graphrag Developer Challenge

Bullaki; Cagliari; In loco; EUR 30.000 - 50.000; Part-time

Descrizione del lavoro

A technology consulting company in Cagliari is looking for a Senior RAG Systems Developer to create a prototype for legal document processing. This contract role focuses on developing a GraphRAG system, must be proficient in Python 3.12, and will require strong skills in knowledge-graph construction and multi-hop reasoning. The position includes a technical evaluation phase that could lead to long-term employment.

Competenze

• Proven experience in building high-accuracy prototypes for legal document reasoning.
• Expertise in multi-hop reasoning and explainability.
• Ability to work independently without UI or provided API keys.

Mansioni

• Build a GraphRAG system for legal reasoning.
• Implement ingest() to create a knowledge graph from Markdown documents.
• Develop query() to answer questions using retrieved sources.

Conoscenze

Graph-based retrieval expertise

Python 3.12

Knowledge-graph construction

Parallel execution support

Testing for correctness and performance

Descrizione del lavoro

GraphRAG Developer Challenge – Legal Document Processing (Prototype)

Role Details

• Role: Senior RAG Systems Developer (Contract / Freelance)
• Compensation: $600 — paid only if you pass (95% benchmark)
• Timeline: 3–5 days from materials receipt to live demo
• Purpose: Technical evaluation for potential long-term hire
• Frontend / UI: None (backend prototype only)

Objective

We’re seeking an expert in graph-based retrieval (GraphRAG) to build a high-accuracy prototype for legal document reasoning. This is a paid technical test that may lead to a long-term position. The goal is a true GraphRAG system featuring explicit knowledge-graph construction and traversal, multi-hop reasoning, agentic orchestration, and strong focus on retrieval accuracy and explainability.

Materials Provided

• /docs/ → Pre-processed Markdown legal documents with metadata
• /sample_questions.json → Example question format
• /sample_answers_rag.json → Example answer format

Download materials:

(Benchmark uses unseen questions.)

Deliverables

Implement two functions in Python 3.12 (Poetry project):

• def ingest(document_paths: List[str]) -> None: 'Ingest Markdown docs and build the knowledge graph.'
• def query(questions: List[str]) -> List[str]: 'Return answers with Vancouver-style citations grounded in retrieved sources.'

Requirements: No UI, no API keys provided. Any stack may be used. query(...) must support parallel execution (~400 questions in ≤60 min) and show a progress indicator. Test thoroughly for correctness and performance before the demo.

Live Demo

In a 60-minute live session you will:

• Receive ~400 unseen questions.
• Run query(...) to produce answers.json.
• Explain your architecture: how the graph is built, traversed, and used to generate grounded answers.
• Only the developer(s) who wrote the code may present.

Evaluation

Passing requires an overall score above 95%, measured by (LLM as a judge): Faithfulness (grounded, no hallucinations), Relevance (retrieval matches intent), Completeness (covers key legal points), and Clarity (structured, legally coherent writing).