⚡ Core Algorithm

Precision Retrieval with Beam Search

Aevum's proprietary beam search algorithm explores multiple knowledge pathways simultaneously, ensuring you receive the most relevant, context-aware results from our 2.4M+ article database.

// Beam Search Visualization

Beam Width (k)

k = 2

Depth

depth = 3

Query Parsing Candidate Generation Scoring Selection

Process

How Beam Search Works

🔍

Query Tokenization

Your query is broken into semantic tokens using Aevum's NLP engine, identifying entities, intent, and context.

🌿

Branch Exploration

The algorithm generates top-k candidate paths through our knowledge graph, exploring related concepts and synonyms.

📊

Scoring & Ranking

Each path is scored based on relevance, authority, recency, and your personalization profile.

✂️

Pruning

Lower-scoring branches are pruned, keeping only the most promising paths to optimize speed and accuracy.

🎯

Result Synthesis

The top beam paths are merged into a unified, coherent response with source citations and knowledge cards.

Implementation

Under the Hood

beam_search.py

# Aevum Beam Search Core Logic
def beam_search(query, knowledge_graph, k=2, max_depth=5):
    beams = [("", 1.0)]  # (path, score)
    
    for depth in range(max_depth):
        candidates = []
        for path, score in beams:
            nodes = knowledge_graph.expand(path, query)
            for node in nodes:
                new_score = calculate_score(node, query)
                candidates.append((path + [node], score * new_score))
        
        # Keep top-k beams
        beams = sorted(candidates, key=lambda x: x[1], reverse=True)[0:k]
    
    return synthesize_results(beams)

<50ms

Average Latency

99.2%

Precision@1

4.8x

Faster than DFS

2.4M

Nodes Indexed

Experience the Power of Beam Search

Try our advanced search engine and see how Aevum retrieves knowledge.

Try Search Now → View API Docs