Loading patterns…

🔧

Dynamic Context Assembly(DCA)

Intelligent, real-time composition of context from multiple sources based on query analysis and relevance scoring

Complexity: highContext Orchestration

In 30 seconds

What: Analyzes incoming queries to identify relevant domains, then assembles context from multiple sources ranked by relevance scores before generating responses.
When to use: Complex queries spanning multiple domains where pulling all available context would exceed token limits or degrade answer quality.
Watch out: Relevance scoring mistakes cause critical information to be excluded, leaving blind spots the model doesn't know exist.

Ask the AI expert about this pattern

Opens the assistant with your question prefilled. You review it before sending.

Get this catalog as an MCP server Tune a prompt for this pattern

Intelligent, real-time composition of context from multiple sources based on query analysis and relevance scoring

Multi-source context fusion
Real-time relevance scoring
Adaptive context selection
Contextual routing mechanisms
Quality-based prioritization
Dynamic context window optimization

The Agent Architect

One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.

Weekly email, one-click unsubscribe. We only use your address to send the briefing.

References

The papers, specifications, and repositories this pattern is based on.

Retrieval-Augmented Generation (Lewis et al., 2020)
Self-RAG and feedback-aware retrieval (Asai et al., 2023)
HyDE: Hypothetical Document Embeddings (Gao et al., 2022)
Fusion-in-Decoder and reranking techniques for open-domain QA
Mixture-of-Experts routing and gating for conditional computation

From the engineer behind this catalog

Get your agent architecture reviewed

This page documents one pattern. Your system runs dozens, and most failures live in how they fit together. Have the whole design reviewed against the 288 patterns in this catalog: architecture, reliability, evaluation and cost, every finding mapped to the pattern that fixes it.

€750 instead of €1,500, one week, written report and walkthrough call, until 30 September

See the review

Patterns

Loading patterns…

Design Patterns & Techniques

Loading patterns…

🔧

Dynamic Context Assembly(DCA)

Intelligent, real-time composition of context from multiple sources based on query analysis and relevance scoring

Complexity: highContext Orchestration

In 30 seconds

What: Analyzes incoming queries to identify relevant domains, then assembles context from multiple sources ranked by relevance scores before generating responses.
When to use: Complex queries spanning multiple domains where pulling all available context would exceed token limits or degrade answer quality.
Watch out: Relevance scoring mistakes cause critical information to be excluded, leaving blind spots the model doesn't know exist.

Ask the AI expert about this pattern

Opens the assistant with your question prefilled. You review it before sending.

Get this catalog as an MCP server Tune a prompt for this pattern

Core Mechanism

Dynamic Context Assembly composes the optimal context window at runtime by analyzing the query, scoring candidate sources (relevance, quality, freshness, diversity, risk), allocating a token budget across sources, and assembling a deduplicated, compressed, and ordered context for generation. The assembly adapts during generation using utilization signals and feedback.

🧭

Analysis

Intent & domain detection

📊

Scoring

Relevance × quality × freshness

🧱

Budgeting

Token allocation per source

🧹

Normalization

Dedup + compression + ordering

Workflow / Steps

Parse query for domains, intent, complexity, time-sensitivity, and constraints.
Enumerate candidate sources (KB, vector DB, docs, APIs, logs) with metadata and costs.
Score candidates on relevance, quality, freshness, coverage, risk; calibrate scores.
Allocate token budget across sources; set per-source caps and ordering policy.
Retrieve top chunks; deduplicate semantically; compress while preserving salience.
Assemble context with clear sectioning and provenance; include task-specific hints.
Generate; monitor utilization and citations; adapt weights or re-pull on gaps.
Log features, decisions, budgets, and outcomes; feed back to improve scoring.

Best Practices

Maintain a typed source registry with SLAs, costs, freshness, and compliance tags.

Use calibrated scorers (BM25/embedding + rerankers) and combine with quality/freshness weights.

Budget first: decide token allocation and per-source caps before retrieval.

Normalize: deduplicate semantically, remove boilerplate, and compress with salience-aware summaries.

Keep provenance (source, chunk id, timestamp) and surface citations in outputs.

Cache frequent queries and intermediate retrievals; invalidate with freshness rules.

Introduce diversity constraints to avoid near-duplicate evidence; prefer complementary sources.

Guardrails: PII/compliance filters pre- and post-assembly; enforce allow/deny by region/role.

Online evaluation loop using labeled queries and rubric-based graders; close the feedback cycle.

Shadow new assembly policies before rollout; compare cost/quality vs. current baseline.

When NOT to Use

Simple, single-source tasks where retrieval + direct prompting suffices.
Strict real-time constraints that cannot afford retrieval and reranking overhead.
Scenarios with tight compliance boundaries prohibiting cross-source mixing.
When high determinism is required and dynamic inputs may reduce reproducibility.
Very small contexts where assembly overhead outweighs benefit.

Common Pitfalls

Overfilling the window with redundant or low-salience chunks (“context stuffing”).
Ignoring deduplication and normalization → diluted signals and higher cost.
Using stale content due to missing freshness policies or cache invalidation.
Uncalibrated scores; heavy reliance on a single similarity metric.
No provenance or citations; harder to audit, recover, or debug hallucinations.
Unbounded token growth without strict per-source caps and global budgets.

Key Features

Multi-source fusion with calibrated relevance/quality/freshness weighting

Budgeted token allocation and ordering policy

Semantic deduplication, normalization, and compression

Adaptive assembly with utilization feedback during generation

Caching, prewarming, and freshness-aware invalidation

Provenance, citations, and explainable assembly rationale

Compliance and safety filtering pre/post assembly

KPIs / Success Metrics

Relevance coverage and answer quality vs. evaluator/oracle.
Source utilization efficiency and marginal benefit per token.
Novelty/diversity score; reduction in redundancy.
Token efficiency (quality per 1K tokens) and total cost per answer.
Latency overhead of assembly (P50/P95) and generation completion rate.
Hallucination rate and citation accuracy/consistency.
Freshness adherence and SLA compliance for time-sensitive queries.

Token / Resource Usage

Total ≈ query analysis + scoring + retrieval + assembled context + generation.
Use small, fast models for scoring/reranking; reserve large models for generation.
Set strict per-source caps and global budgets; prefer compression over dropping sources.
Exploit caching and prefetch for hot queries; stream retrieval to overlap with scoring.
Prefer rerankers to reduce k before assembly; adjust chunk size for salience density.
Throttle concurrency and apply backpressure; avoid explosive fan-out across sources.

Best Use Cases

Multi-domain question answering and enterprise knowledge synthesis.
Research assistants comparing papers, standards, and reports across sources.
Analyst copilots needing balanced evidence and citations.
Customer support with product docs, tickets, and policy KBs.
Compliance/policy answering requiring provenance and freshness.

References & Further Reading

Academic Papers

Retrieval-Augmented Generation (Lewis et al., 2020)
Self-RAG and feedback-aware retrieval (Asai et al., 2023)
HyDE: Hypothetical Document Embeddings (Gao et al., 2022)
Fusion-in-Decoder and reranking techniques for open-domain QA
Mixture-of-Experts routing and gating for conditional computation

Implementation Guides

Multi-source RAG orchestration and context assembly patterns
LangGraph/LangChain routing + RAG pipelines with rerankers
LlamaIndex QueryEngine + Node PostProcessors for compression
DSPy pipelines with retrieval, rerank, and adaptive prompting

Tools & Libraries

LangGraph, LangChain, LlamaIndex, DSPy, Haystack
Vector DBs: FAISS, Pinecone, Weaviate, Milvus, Vectara
Rerankers: Cohere Rerank, ColBERT, cross-encoders
Compression: token/semantic compressors and TL;DR summarizers

Community & Discussions

RAG surveys and best practices (2023–2025)
LlamaIndex and LangChain community forums
LLM ops groups on evaluation, routing, and context optimization

Intelligent, real-time composition of context from multiple sources based on query analysis and relevance scoring

Multi-source context fusion
Real-time relevance scoring
Adaptive context selection
Contextual routing mechanisms
Quality-based prioritization
Dynamic context window optimization

The Agent Architect

One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.

Weekly email, one-click unsubscribe. We only use your address to send the briefing.

References

The papers, specifications, and repositories this pattern is based on.

Retrieval-Augmented Generation (Lewis et al., 2020)
Self-RAG and feedback-aware retrieval (Asai et al., 2023)
HyDE: Hypothetical Document Embeddings (Gao et al., 2022)
Fusion-in-Decoder and reranking techniques for open-domain QA
Mixture-of-Experts routing and gating for conditional computation

From the engineer behind this catalog

Get your agent architecture reviewed

€750 instead of €1,500, one week, written report and walkthrough call, until 30 September

See the review

Loading patterns…

Dynamic Context Assembly(DCA)

In 30 seconds

Ask the AI expert about this pattern

Dynamic Context Assembly: Overview

The Agent Architect

References

Get your agent architecture reviewed

Patterns

Loading patterns…

Design Patterns & Techniques

Loading patterns…

Dynamic Context Assembly(DCA)

In 30 seconds

Ask the AI expert about this pattern

Core Mechanism

Workflow / Steps

Best Practices

When NOT to Use

Common Pitfalls

Key Features

KPIs / Success Metrics

Token / Resource Usage

Best Use Cases

References & Further Reading

Academic Papers

Implementation Guides

Tools & Libraries

Community & Discussions

Dynamic Context Assembly: Overview

The Agent Architect

References

Get your agent architecture reviewed

Patterns

Design Patterns & Techniques

Loading patterns…

Design Patterns & Techniques

Core Mechanism

Workflow / Steps

Best Practices

When NOT to Use

Common Pitfalls

Key Features

KPIs / Success Metrics

Token / Resource Usage

Best Use Cases

References & Further Reading

Academic Papers

Implementation Guides

Tools & Libraries

Community & Discussions

Design Patterns & Techniques