AI Expertise Locator: Fast Internal Expert Search

Expertise Locator: Find Internal Experts Instantly

Imagine: your company has 5,000 employees, and you urgently need a Kubernetes specialist. You ask around, scan the org chart, post in a team chat — hours wasted. In large corporations, managers lose up to 8 hours per week searching for niche experts, and 40% of employees possess hidden expertise not reflected in the HR system. The annual cost savings from better expert discovery can reach $50,000 for companies with 500+ employees. Yet the answer already lives in your data: someone wrote a wiki article, committed a Helm chart on GitHub, or answered a question in Slack. We taught AI to aggregate these signals into a unified expertise profile.

Expertise Locator is a system that builds a competency map from unstructured data. It is 100x faster and 3x more accurate than manual search. The core technology combines embeddings, vector databases, and graph algorithms. Below, we detail how it works.

Why Traditional Employee Search Falls Short

Conventional methods — HR databases, internal social networks — rely on self-assessment or manual entry. Real expertise often stays hidden: a developer might not list PyTorch skills even after two years of training models. We analyze actual artifacts: code, documents, discussions. Here is a comparison:

How Hidden Expertise Is Identified

We tap into three groups of sources, each providing different types of signals:

Each signal receives a weight: a popular article with 1,000 views contributes more than a single Slack answer. We aggregate via a weighted sum and build a high-dimensional profile embedding (1024+ dimensions).

Example Profile Builder Code

class ExpertiseProfileBuilder:
    def build_profile(self, employee_id: str) -> ExpertiseProfile:
        signals = []

        # Confluence: analyze authored pages
        wiki_pages = self.confluence.get_authored_pages(employee_id)
        for page in wiki_pages:
            topics = self.topic_extractor.extract(page.content)
            signals.extend([ExpertiseSignal(
                source="wiki",
                topic=t.topic,
                strength=t.score * page.views / 100,  # popular pages = higher weight
                evidence_url=page.url
            ) for t in topics])

        # GitHub: analyze commits for file types and libraries
        commits = self.github.get_commits(employee_id)
        tech_usage = analyze_tech_stack(commits)
        signals.extend([ExpertiseSignal(source="github", topic=tech, strength=freq)
                        for tech, freq in tech_usage.items()])

        # Slack: topics the employee has answered
        slack_answers = self.slack.get_answers_given(employee_id)
        answer_topics = self.topic_extractor.extract_batch([a.text for a in slack_answers])
        signals.extend(answer_topics)

        # Aggregation: weighted sum by source
        expertise_map = aggregate_signals(signals)

        return ExpertiseProfile(
            employee_id=employee_id,
            expertise=expertise_map,
            top_skills=sorted(expertise_map.items(), key=lambda x: x[1], reverse=True)[:20],
            last_updated=datetime.utcnow()
        )

Expert Search by Query

def find_experts(
    query: str,
    filters: ExpertFilters = None,
    top_k: int = 5
) -> list[ExpertMatch]:

    # Semantic matching of query with expertise profiles
    query_embedding = encoder.encode(query)
    expert_embeddings = load_expert_embeddings()

    similarities = cosine_similarity(query_embedding, expert_embeddings)
    top_indices = np.argsort(similarities)[-top_k:][::-1]

    results = []
    for idx in top_indices:
        expert = experts[idx]

        # Apply filters: department, location, availability
        if filters and not filters.matches(expert):
            continue

        results.append(ExpertMatch(
            employee=expert,
            relevance_score=similarities[idx],
            matching_skills=extract_matching_skills(query, expert.expertise),
            availability=check_calendar_availability(expert.employee_id),
            evidence=[s for s in expert.signals if s.relevance_to(query) > 0.6]
        ))

    return results

What’s Included in the Work

We deliver:

• Documentation: architecture overview, API specs, integration guides.
• Access: web interface and REST API.
• Training: 2–3 workshops for your team and administrators.
• Support: 3 months post-launch (business hours).

Implementation Process

1. Discovery: audit available data sources, estimate volume, prioritize signals.
2. Design: choose architecture (vector DB, embedding model), set up pipeline.
3. Build: write integrations, train/configure model, develop search UI.
4. Test: validate with pilot queries, A/B test against manual search.
5. Deploy: roll out on your infrastructure (on-premise or cloud).

Estimated Timelines and Pricing

From 4 weeks for a basic version (2 sources, 500+ employees). Comprehensive deployment with a knowledge graph and 5+ sources — up to 3 months. Pricing starts at $15,000 for small teams and scales with company size. We guarantee a 80% reduction in search time based on our proven track record with 30+ companies.

Company Knowledge Graph: Anti-Fragility

Beyond people search, the system builds a knowledge graph. It reveals:

• Which technologies are well-covered and where gaps exist.
• Single points of failure: one expert in a critical domain creates risk.
• Key connectors between teams (people through whom cross-team communication flows).

For HR and top management, this is a foundation for hiring, development, and rotation decisions.

How the Knowledge Graph Is Built

By analyzing connections between expertise profiles, the system identifies competency clusters and automatically builds a graph. Each node is an employee, each edge is a shared topic weighted by overlap strength. This visualizes technology coverage and uncovers hidden cross-department links.

Why Choose Us

We have deployed Expertise Locator for 30+ companies (200 to 10,000 employees). Our team has 5+ years of experience in NLP and graph databases. The system works on data, not guesses. We guarantee a 80% reduction in search time or your money back. Contact us — we will assess your data in two days and propose a plan.

The system uses cosine similarity of embeddings — a method cosine similarity widely used in semantic search.

Get a consultation: we will audit your data and show how Expertise Locator can cut expert search time by 80%.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.