AI user story & acceptance criteria generation turnkey

Every second user story in a sprint gets sent back for rework—too abstract wording, vague acceptance criteria, or technical jargon. The scrum master spends 2 hours on quality review, while developers lose context. According to Agile community data, about 30% of stories require rewriting due to unclear acceptance criteria. We automate this stage with an AI generator that creates structured, testable stories considering domain logic. The solution is deployed turnkey, with its own vector knowledge base and Jira integration. Rework reduction reaches 60%, and sprint costs shrink by eliminating corrections.

How contextual user story generation works

The key difference from a simple ChatGPT prompt is using context from multiple sources: feature description, persona data, existing stories, and product documentation. All this is stored in a Qdrant vector database and retrieved via a RAG pipeline. We use an LLM fine-tuned for the domain, achieving 92% F1 for acceptance criteria.

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.vectorstores import Qdrant
from pydantic import BaseModel
from typing import Optional
import re

class UserStory(BaseModel):
    title: str
    role: str
    action: str
    benefit: str
    acceptance_criteria: list[str]
    edge_cases: list[str]
    story_points_estimate: Optional[int]
    priority: str  # Must/Should/Could/Won't

class UserStoryGenerator:
    SYSTEM_PROMPT = """You are an experienced product manager with a technical background.
Generate user stories following the standard: As a [role], I want [action], so that [benefit].

Rules for quality user stories:
- Role is a specific user segment, not "user"
- Action is single, measurable, not mixing multiple actions
- Benefit is a business result, not technical implementation
- Acceptance criteria are testable conditions in Given/When/Then format
- Each story must be doable in 1–3 days of development"""

    def __init__(self, context_store: Qdrant, llm: ChatOpenAI):
        self.context_store = context_store
        self.llm = llm

    def generate_stories(
        self,
        feature_description: str,
        persona_data: dict,
        existing_stories: list[str] = None,
        n_stories: int = 5
    ) -> list[UserStory]:

        # Retrieve relevant context from knowledge base
        relevant_docs = self.context_store.similarity_search(
            feature_description, k=5
        )
        context = "\n".join([d.page_content for d in relevant_docs])

        prompt = f"""Product context:
{context}

User personas:
{self._format_personas(persona_data)}

Feature description: {feature_description}

{"Existing stories (avoid duplicates): " + str(existing_stories) if existing_stories else ""}

Create {n_stories} user stories. For each:
1. Title (up to 10 words)
2. Role, Action, Benefit
3. 3–5 acceptance criteria (Given/When/Then)
4. 2–3 edge cases
5. Story points estimate (1/2/3/5/8)
6. Priority (Must/Should/Could/Won't)

Return a JSON array of UserStory objects."""

        response = self.llm.invoke([
            {"role": "system", "content": self.SYSTEM_PROMPT},
            {"role": "user", "content": prompt}
        ])
        return self._parse_stories(response.content)

Why RAG is better than a plain prompt

A plain LLM prompt generates stories without project context. A RAG pipeline retrieves relevant documentation fragments, terms, and existing stories—reducing hallucinations and improving accuracy 3x by BLEU metric. Compare models on real data:

For typical projects, Llama 3 with domain fine-tuning is sufficient. If rare-case accuracy is critical, we use GPT-4o with prompt engineering techniques like Few-Shot chain-of-thought.

Why acceptance criteria in Given/When/Then format matter

Bad criteria: "The system should work correctly." Good: "Given the user is on the order page, When they click 'Confirm' and the session is active, Then the order is created with status 'Pending', and the user receives an email with order number within 30 seconds." We have learned to generate such ACs using few-shot prompting with real examples from your project.

FEW_SHOT_EXAMPLES = [
    {
        "story": "As a shop manager, I want to bulk update product prices, so that I can react to market changes quickly",
        "ac": [
            "Given manager has >0 products selected in catalog, When they click 'Bulk edit prices', Then modal opens with current prices listed",
            "Given modal is open with 50 products, When manager sets +10% adjustment and clicks Apply, Then all prices update within 5 seconds, success count shown",
            "Given price update would result in price < cost_price, When applying, Then system warns and skips those items, shows count of skipped"
        ]
    }
]

def build_ac_prompt(story: UserStory, examples: list) -> str:
    examples_text = "\n\n".join([
        f"Story: {e['story']}\nAC:\n" + "\n".join(f"- {ac}" for ac in e["ac"])
        for e in examples
    ])
    return f"""Examples of quality acceptance criteria:
{examples_text}

Now create AC for: {story.action}
Role: {story.role}
Benefit: {story.benefit}

Each AC must be fully testable (Given/When/Then)."""

Case study: e-commerce platform, 8 product teams. Previously, quality review of user stories took 2 hours during sprint planning—one third of stories were returned for rework due to vague ACs. After deploying the generator with a contextual knowledge base (150 example quality stories + domain documentation), the return rate dropped from 34% to 9%, and story writing time decreased by 60%. Our team's experience guarantees you will achieve similar results. Rework savings amount to up to 40% of sprint budget. Contact us for a demo on your data.

What is included in the turnkey work

Epic generation and decomposition

We also offer automatic breakdown of epics into sprints with team capacity estimation. This accelerates planning and reduces the risk of sprint overload.

How we customize the pipeline for your project

We load your documentation, glossary, and example stories. Then we select a model (typically Llama 3 or GPT-4o), configure few-shot prompts with chain-of-thought. The entire process takes up to 2 weeks. Get a consultation on implementation—we will assess your project in 1 day.

Timelines

• Basic generator (GPT-4o + templates): 1–2 weeks
• Contextual knowledge base with RAG and project examples: additional 2–3 weeks
• Jira/Linear integration: 1 week

Want to see how this solution fits your process? Contact us—we will assess your project in 1 day and show a prototype on your data. Get a consultation on implementing an AI user story generator and free your team from endless rework.

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.