AI-driven test failure analysis and error classification system

Note: When dozens of tests fail simultaneously in CI, root cause analysis takes hours. Engineers scroll through logs trying to understand what broke: infrastructure, regression, or a flaky test. We trained a model to automatically group failures by real cause and route tickets to the responsible developer. Our AI system processes stack traces 5–10 times faster than manual analysis, reducing time from failure to fix from hours to minutes. Implementing this approach saves a team up to 80 person-hours per month on CI failure triage.

The system doesn't just find errors — it prioritizes defects and automatically routes notifications. Instead of alerting the entire team, only the person who can actually fix the problem receives the notification.

How does error classification work?

The pipeline consists of four stages: parsing, classification, clustering, and routing. Input: raw test logs. Output: structured reports with priority and owner.

Parsing — supports JUnit XML, Allure, pytest JSON, Cypress Mochawesome. Normalizes into a unified schema: {test_id, status, duration, error_message, stack_trace, timestamp}.

Classification — a multi-label classifier based on fine-tuned CodeBERT, trained on labeled stack traces. Output classes:

• INFRASTRUCTURE — timeouts, connection refused, OOM
• REGRESSION — test passed, broke after a commit
• FLAKY — unstable (passes with retry)
• TEST_BUG — error in the test itself, not in the code
• NEW_BUG — real defect in the product
• ENVIRONMENT — problem with test environment

Clustering — stack trace embeddings via CodeBERT → HDBSCAN clustering. Tests in the same cluster likely share the same root cause.

from sentence_transformers import SentenceTransformer
import hdbscan

class ErrorClusterer:
    def __init__(self):
        self.encoder = SentenceTransformer('microsoft/codebert-base')
        self.clusterer = hdbscan.HDBSCAN(
            min_cluster_size=3,
            metric='cosine',
            cluster_selection_method='eom'
        )

    def cluster_failures(self, failures: list[dict]) -> list[int]:
        texts = [f['error_message'] + '\n' + f['stack_trace'][:500]
                 for f in failures]
        embeddings = self.encoder.encode(texts, batch_size=32)
        return self.clusterer.fit_predict(embeddings)

Why use AI test analysis?

Manual failure triage is time-consuming and error-prone. Fine-tuned CodeBERT processes stack traces 5 times faster than a human, and clustering reveals cascading failures that appear as dozens of independent errors. We guarantee: after deployment, you will halve your CI failure triage time. Our experience shows that within 2–3 weeks the system reaches target quality metrics.

How HDBSCAN clustering reveals cascading failures?

Cascading failures — one breakdown (e.g., database outage) causes dozens of test failures. Clustering groups them by stack trace and error message similarity. Even if errors appear different (connection timeout, NULL pointer, 500 error), HDBSCAN merges them into one cluster if embeddings are close enough. This immediately reveals that all failures stem from a single issue.

Flaky test detection

Flaky tests form a separate category. The system analyzes the history of the last 30 runs for each test and computes a flakiness score:

flakiness_score = std(pass_rate_per_day) * frequency_of_status_changes

A test with flakiness_score > 0.3 is marked as flaky and automatically quarantined — it continues running for statistics but does not block CI.

Notification routing

The system determines the owner of the failing test via git blame on the test file and files from the stack trace. Notifications are sent not to the whole team, but:

• REGRESSION/NEW_BUG → author of the last commit on affected files + team lead
• FLAKY → test owner from git blame
• INFRASTRUCTURE → DevOps channel
• TEST_BUG → QA responsible for that test file

Classification quality metrics

Training is performed on a labeled dataset from the project history. Typical metrics after 2–3 weeks of data accumulation:

What is included in the work?

• Audit of the current CI process and report formats
• Parsing and log normalization setup
• Classifier training on your historical data
• Integration with GitLab, GitHub Actions, or Jenkins
• Dashboard (Grafana) and webhook notifications deployment
• Documentation and team training

Deployment process

1. Analytics — metric collection, defining error types
2. Design — stack selection, pipeline configuration
3. Implementation — writing parsers, model training
4. Testing — A/B comparison with manual analysis
5. Deployment — staging, then production

Estimated timelines

Pricing is calculated individually after an audit. We will estimate your project in one business day — contact us for a consultation. Get a demo access to the system and verify its effectiveness on your own 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.