Self-Healing Tests: Auto-Fixing Broken Tests with AI

Your team spends 20–40% of its time maintaining the test suite — not writing new tests, but fixing broken ones. The culprits: UI changes (locator no longer finds the element), API changes (response structure altered), or business logic changes (outdated assertion). Self-Healing Tests is an ML layer that detects the cause of failure and automatically applies a fix without human intervention. It’s a practical application of AI for testing and ML for QA. Budget savings on test maintenance can reach 50%, and flaky tests drop by up to 70%. Contact us to get a project estimate.

Compare: manually fixing one flaky test takes an average of 2–4 hours. Self-Healing reduces that to 5 minutes for reviewing the proposed fix — a 24–48x speedup.

How Self-Healing Tests Works

The system operates in two modes: proactive — preemptively updating locators when new frontend is deployed, and reactive — fixing after a CI/CD failure.

Core system consists of three modules:

• Failure Classifier — an NLP model (fine-tuned DistilBERT) that classifies failure type from stack trace: ElementNotFound, AssertionError, TimeoutError, NetworkError.
• Selector Healer — for ElementNotFound, finds an alternative locator via DOM analysis; trained on pairs (old locator → new locator) from commit history.
• Assertion Fixer — for AssertionError, compares actual and expected values, identifies the change pattern (numeric drift, string format change, JSON structural change) and suggests an updated assertion.

class SelfHealingRunner:
    def __init__(self, model_path: str):
        self.classifier = FailureClassifier.load(model_path)
        self.healer = SelectorHealer()
        self.assertion_fixer = AssertionFixer()

    def run_with_healing(self, test_fn, max_retries: int = 2):
        for attempt in range(max_retries + 1):
            try:
                return test_fn()
            except Exception as e:
                if attempt == max_retries:
                    raise
                failure_type = self.classifier.predict(str(e))
                if failure_type == "ElementNotFound":
                    self.healer.apply_fix(e)
                elif failure_type == "AssertionError":
                    self.assertion_fixer.suggest(e)

What Selector Healing Delivers

For Selenium or Playwright tests, the main source of flakiness is fragile CSS selectors like #app > div:nth-child(3) > button. After a layout change, such a locator fails.

Recovery algorithm:

1. Parse the current DOM at the moment of failure.
2. Extract features of the lost element from the test source: tag type, text content, aria-label, sibling elements.
3. Build an element embedding (features → vector via trained encoder).
4. Find the nearest element in the current DOM by cosine similarity.
5. Generate a new locator: prefer data-testid, then aria-label, then XPath with text().

Recovery accuracy on a test dataset (5000 pairs): 87% correct fixes. Self-Healing speeds up locator repair by 24–48x compared to manual search. We guarantee that after a two-week report-only mode, the system is ready for auto-fixes.

How We Integrate into CI/CD

# .github/workflows/tests.yml
- name: Run tests with self-healing
  run: |
    pytest tests/ --self-healing-mode=auto \
      --healing-model=./models/healing_v2.pkl \
      --max-healing-retries=2 \
      --healing-report=artifacts/healing_report.json

After each healing event, the system creates a Pull Request with the suggested fix — the engineer only does code review instead of debugging from scratch. On our projects with 500+ e2e tests, auto-healing closes 60–70% of failures without QA involvement, saving over 100 person-hours per month. For Playwright, self-healing is fully implemented; for Selenium, auto-repair works through the Selector Healer. Our experience across 10+ projects confirms consistent results.

Supported Frameworks and Technologies

Implementation Stages

1. Audit test base: analyze failure frequency by type, identify most flaky tests.
2. Collect dataset from CI history — pairs (failed test, fix commit).
3. Train Failure Classifier and Selector Healer on your project.
4. Integrate into CI/CD pipeline with report-only mode for the first 2 weeks.
5. Switch to auto-fix mode with confidence threshold > 0.85.

What’s Included

• Full audit of test infrastructure and failure history.
• Training custom models on your project data.
• CI/CD integration (GitHub Actions, GitLab CI, Jenkins).
• Setup and operation documentation.
• Team training (2-hour webinar).
• 3 months of post-implementation support.

After implementation, failed tests drop by 70% — confirmed by our clients’ experience. Self-Healing automates regression testing and reduces flaky tests.

When to Order Implementation

If your team spends more than 20% of its time on test maintenance and flaky tests increase every sprint — contact us. We’ll assess your project in 1–2 days and propose a solution. Order implementation and get a free consultation on your test base.

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.