AI-Powered Knowledge Base Curation: Audit, Recommendations, Dashboard

AI-Powered Knowledge Base Curation and Updates

We often see that a neglected knowledge base is worse than no knowledge base at all. It creates an illusion of order but is actually full of outdated instructions, duplicate articles, and dead links. Manual curation of 500+ articles is unrealistic — the editor simply lacks time to check each one systematically.

Our AI curation system doesn't replace experts, but it automatically detects issues, prioritizes reviews, and suggests specific edits — the editor works with a ready-made review list, not raw content. Over 5 years of work, we've processed knowledge bases ranging from 200 to 10,000 articles and guarantee improved freshness and coverage metrics.

How AI Detects Outdated Content

The system analyzes each article across several parameters. First, it checks the last update date and compares it with the topic: software version articles become outdated faster than conceptual materials. Second, LLM evaluates the content for references to obsolete versions, product names, or team names. For example, if an article mentions v1.2 and the current version is v3.0, a flag is raised. Third, it checks internal links: broken links are sent to a list for correction. For embeddings, we use SentenceTransformer, and for LLM orchestration, LangChain.

Five Types of Problems AI Detects

1. Outdated content — software versions, team names, links to non-existent processes.
2. Duplication — semantically similar articles under different URLs.
3. Gaps — questions users frequently ask but no article exists.
4. Low quality — articles without acceptance criteria, examples, or with vague wording.
5. Broken relationships — internal links lead to non-existent pages.

from langchain_openai import ChatOpenAI
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
from datetime import datetime, timedelta

class KnowledgeBaseAuditor:
    def __init__(self, confluence_client, llm: ChatOpenAI, embedder: SentenceTransformer):
        self.confluence = confluence_client
        self.llm = llm
        self.embedder = embedder

    def find_duplicates(self, articles: list[dict], threshold: float = 0.88) -> list[tuple]:
        """Finds semantically duplicate articles"""
        texts = [f"{a['title']} {a['body'][:500]}" for a in articles]
        embeddings = self.embedder.encode(texts, batch_size=32, show_progress_bar=True)

        # Pairwise comparison — only upper triangle of the matrix
        sim_matrix = cosine_similarity(embeddings)
        np.fill_diagonal(sim_matrix, 0)

        duplicates = []
        for i in range(len(articles)):
            for j in range(i + 1, len(articles)):
                if sim_matrix[i][j] >= threshold:
                    duplicates.append((
                        articles[i]["id"], articles[i]["title"],
                        articles[j]["id"], articles[j]["title"],
                        round(float(sim_matrix[i][j]), 3)
                    ))

        return sorted(duplicates, key=lambda x: x[4], reverse=True)

    async def check_staleness(self, article: dict) -> dict:
        """Evaluates article staleness via LLM"""
        last_updated = datetime.fromisoformat(article["last_updated"])
        age_days = (datetime.now() - last_updated).days

        prompt = f"""Evaluate the freshness of a technical article.

Title: {article['title']}
Content (first 1000 characters):
{article['body'][:1000]}

Last updated: {age_days} days ago

Check:
1. Are specific software/tool versions mentioned? Are they current?
2. Are there obsolete terms (e.g., old team or product names)?
3. Do links seem current?
4. Overall assessment: current/needs_review/outdated

Return JSON: {staleness_level, reasons: [], suggested_action}"""

        result = await self.llm.ainvoke(prompt)
        return {"article_id": article["id"], **eval(result.content)}

    def find_content_gaps(
        self,
        support_queries: list[str],
        articles: list[dict]
    ) -> list[dict]:
        """Finds questions without corresponding articles"""
        # Index existing articles
        article_embs = self.embedder.encode(
            [a["title"] + " " + a["body"][:200] for a in articles]
        )

        gaps = []
        for query in support_queries:
            query_emb = self.embedder.encode([query])
            similarities = cosine_similarity(query_emb, article_embs)[0]
            max_sim = np.max(similarities)

            if max_sim < 0.65:  # no sufficiently similar article
                gaps.append({
                    "query": query,
                    "best_match_score": round(float(max_sim), 3),
                    "best_match_article": articles[np.argmax(similarities)]["title"]
                })

        return sorted(gaps, key=lambda x: x["best_match_score"])

Automatic Improvement Suggestions

After problem detection, we generate specific enhancements:

    async def suggest_improvements(self, article: dict) -> dict:
        prompt = f"""Review the knowledge base article and suggest concrete improvements.

Article: {article['title']}
Text:
{article['body'][:2000]}

Evaluate and suggest edits for each criterion:
1. Clarity — is the article understandable for a new employee?
2. Completeness — are there examples, specific steps?
3. Structure — are headings, lists needed?
4. Currency — no outdated details?

For each issue: original quote → suggested fix.
Return JSON: {score: 0-10, issues: [{location, problem, suggestion}]}"""

        result = await self.llm.ainvoke(prompt)
        return {"article_id": article["id"], **eval(result.content)}

Knowledge Base Health Dashboard

Why Automated Curation Is More Effective Than Manual

Comparison: a human editor spends on average 15–20 minutes checking one article. For an 800-article base, that's 200–270 hours of continuous work — not counting time for fixes. An AI system processes the same volume in 2–3 hours of analysis and outputs a prioritized list of issues. The editor only needs to approve or reject the suggested edits. Thus, AI reduces curation time by 15–20 times while maintaining quality.

What's Included

• Audit of the current knowledge base: full analysis of articles, detection of duplicates, outdated content, gaps.
• Setting up the AI pipeline: integration with Confluence, Notion, GitBook, or API; model selection, similarity threshold tuning.
• Dashboard with metrics: visualization of Freshness Score, Coverage Rate, Duplication Rate, Quality Score.
• Automatic recommendations: LLM generates specific edits for each problematic article.
• Team training: workshop on using the system, handover of documentation.
• Support guarantee: 3 months of post-implementation support.

Based on our data, after AI curation implementation, companies reduce knowledge base maintenance time by 60–80%.

Process

1. Analytics: collect and structure existing articles, export search logs and support queries.
2. Design: choose architecture (RAG, fine-tuning, or zero-shot), select model and embedder.
3. Implementation: develop audit pipeline, integrate with storage, set up dashboard.
4. Testing: run on test sample, cross-check with expert evaluation, calibrate thresholds.
5. Deployment: launch in production, hand over to team, training.

Timeline and Cost

Timelines depend on the base volume and integration complexity. Audit and problem detection take 1 to 2 weeks. The automatic recommendation system and dashboard take another 3–4 weeks. Cost is calculated individually after assessing your knowledge base and current processes. We guarantee transparent pricing and cost lock-in at the contract stage.

Get a free project assessment. Contact us for a preliminary analysis and metrics dashboard of your knowledge 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.