Prompt Chaining Implementation: Chain-of-Thought Pipelines for LLMs

Implementing Prompt Chaining (Prompt Chains)

Note: when a single LLM call can't handle a complex task—classifying a document, extracting data, validating it, and generating a response—Prompt Chaining comes to the rescue. This is a sequence of calls where the output of the previous step becomes the input of the next. We use this technique for automating document processing, and the results speak for themselves: 71% of tasks are processed without human intervention, extraction accuracy reaches 94%. Prompt Chaining is 20x faster than manual processing: instead of 15 minutes per document, it takes 45 seconds. This approach is based on Chain-of-thought prompting, which improves LLM reasoning through decomposition.

A typical scenario where a single-call pipeline fails is an overloaded context window. When you try to cram all logic into one prompt, the model starts hallucinating, missing fields, or producing unstructured output. A chain splits the task into steps of 500-1000 tokens each, each model works in its own mode, and you get control at every stage. We set temperature, max_tokens, and few-shot examples for each link to guarantee predictable results.

What problems does Prompt Chaining solve?

• Hallucinations with overloaded context: a single prompt with 5000+ tokens loses focus. The chain breaks the task into steps of 500-1000 tokens, each model works in its own mode.
• Inability to validate intermediate results: in a chain, you can check the sanity of extracted data before the next call—this reduces errors by 40-60%.
• Rigid sequence with branching: documents of different types (invoice, contract, complaint) require different extraction schemas. A branching chain automatically selects the right prompt based on classification.
• LLM parallelization: independent steps can be executed simultaneously, reducing p99 latency.
• RAG chain: the chain easily incorporates vector search for context augmentation.

Why is a prompt chain more effective than a single call?

A single complex prompt often leads to overfitting on frequent patterns and ignoring rare ones. A chain allows specializing each step: a classifier with temperature 0, an extractor with few-shot examples, a validator with a strict schema. You get not just an answer but a verifiable pipeline. For example, when processing invoices, we first classify the document type, then extract fields according to an exact schema, then validate the format—and only then pass it to business logic. This approach yields an F1-score of 0.98 on classification and extraction accuracy of 94%.

How we do it: stack, versions, configs

We build chains on the stack:

• Models: OpenAI GPT-4o, Claude 3.5 Sonnet, LLaMA 3 70B (via Together AI).
• Frameworks: Python 3.11 + LangChain for orchestration, Pydantic for data schemas.
• Tools: Weights & Biases for logging, Weaviate for vector search (RAG).
• Deployment: FastAPI + Docker on Kubernetes with GPU inference via vLLM for low p99 latency.

# config/classifier.yaml
model: gpt-4o-2024-08-06
temperature: 0
max_tokens: 50
system_prompt: "Classify the document type in one word: invoice/contract/complaint/inquiry"
validation: true  # checks that the response is one of the four words

Practical case: incoming correspondence processing

Our client—a logistics company—processed 500+ documents daily (invoices, contracts, claims). A single LLM call gave 65% extraction accuracy. We built a 5-step chain:

1. Document type classification (1 LLM call, temperature=0)
2. Data extraction according to type-specific schema (specialized prompt with few-shot examples)
3. Extracted data validation (format check, required fields)
4. Business decision determination (approve/reject/escalate)
5. Response letter generation

Result: autonomous processing of 71% of documents without human intervention, average processing time 45 seconds, extraction accuracy increased to 94%. Classification F1-score: 0.98.

What's included in the work

• Task audit: analyze typical cases and edge cases.
• Chain design: define steps, data schemas, branching and validation points.
• Implementation: write prompts, set temperature, max_tokens, few-shot examples.
• Test integration: unit tests for each step, integration tests for the whole chain.
• Documentation: prompt descriptions, output data schemas, maintenance instructions.
• Team training: show how to add new document types without our involvement.

Work process

1. Analytics (1–2 days): gather requirements, measure current performance.
2. Design (1–3 days): develop chain architecture, select models.
3. Implementation (2–5 days): write code, prompts, validators.
4. Testing (1–2 days): A/B tests on historical data, check edge cases.
5. Deployment (1 day): deploy on your infrastructure or ours (AWS/GCP).
6. Support: one month of free support after launch.

Approximate timelines

Why choose Prompt Chaining?

Prompt Chaining is not just a trendy technique. It's a proven way to consistently obtain high-quality results from LLMs in production. Our engineers have 10+ years of ML experience and guarantee that every chain will pass validation on test data. You get a predictable, controllable, and scalable pipeline that is easy to adapt to new tasks. Get a consultation on your scenario—we'll show how a prompt chain can solve your problem. Contact us for a test run on your data.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.