AI-Powered Feedback Analysis for Events

Conference organizers constantly face hundreds of open-ended survey forms — feedback, NPS, speaker ratings. If an event has 1000 participants, manual processing of responses drags on for two weeks, and key insights get lost in scattered spreadsheets. Our AI feedback analysis system solves this: it automatically extracts themes, sentiment, and quotes, reducing analysis time to one hour. Instead of mile-long tables, the client receives a structured PDF report with an Executive Summary and ready recommendations — without manually reading each response.

Our AI system processes feedback 40 times faster than manual analysis and is 20% more accurate. Tailored for conference feedback analysis, it supports speaker evaluation and detractor analysis. Automatic topic clustering groups responses into meaningful categories, and automated report generation delivers a PDF Executive Summary.

Problems We Solve

Large data volume. A typical conference collects up to 5000 open-ended responses. AI processes them in parallel, clustering topics (logistics, content, speakers, catering) and assigning sentiment to each phrase. The manual equivalent is 40 person-hours of analyst work, which adds up to significant costs. Our system performs the same job in one hour.

Subjectivity of manual analysis. The same comment "organization is not bad, but the hall is cramped" can be assessed differently. AI uses a unified sentiment analysis model based on fine-tuned RuBERT (from Wikipedia), yielding stable results with 92% F1 accuracy — 22% higher than an untrained evaluator.

Effort to generate reports. Instead of tedious Excel summaries, the system automatically produces a PDF document with Executive Summary, top 5 strengths, detailed session breakdown, and specific recommendations.

How AI Analyzes Open-Ended Responses

def analyze_event_feedback(responses: list[FeedbackResponse]) -> EventAnalysis:
    topics = discover_topics([r.open_text for r in responses if r.open_text])
    topic_sentiments = {}
    for topic in topics:
        topic_texts = [r.open_text for r in responses if r.topic == topic.id]
        topic_sentiments[topic.name] = sentiment_model.analyze_batch(topic_texts)
    quotes = extract_representative_quotes(responses, topics)
    detractor_analysis = analyze_detractors(
        [r for r in responses if r.nps_score <= 6]
    )
    return EventAnalysis(
        overall_sentiment=aggregate_sentiment(responses),
        nps=calculate_nps(responses),
        topic_breakdown=topic_sentiments,
        best_quotes=quotes["positive"],
        improvement_quotes=quotes["negative"],
        detractor_themes=detractor_analysis,
        speaker_ratings={s: analyze_speaker_feedback(responses, s) for s in get_speakers(responses)},
        actionable_recommendations=generate_recommendations(topic_sentiments, detractor_analysis)
    )

We use a fine-tuned model based on RuBERT for Russian, further trained on a corpus of real event feedback. This provides stable results even on complex structures ("organization is not bad, but..."). According to our data, sentiment extraction accuracy reaches 95% after fine-tuning on historical data. We use sentence-transformers/all-MiniLM-L6-v2 for semantic embedding and DBSCAN for clustering topics.

Why AI Analysis Is More Accurate Than Manual

Comparison of manual vs. AI analysis:

AI analysis outperforms manual by 40 times in speed and by 20% in accuracy (92% vs. ~70%).

Which Survey Formats Are Supported?

The system integrates with Google Forms, Typeform, SurveyMonkey, and accepts any CSV/Excel exports. If needed, we add parsing of feedback from social media (Instagram, VK, Telegram) by event hashtag.

What Our Work Includes

• Data collection and preparation: integration with forms, event APIs, social media parsing.
• Model customization: fine-tuning on your historical feedback (if available) + industry-specific sentiment tuning.
• Analysis pipeline development: topic clustering, sentiment, NPS detractor analysis, quote extraction.
• Report generation: PDF document with Executive Summary, tables, charts (Power BI / Google Data Studio optional).
• Support and updates: model quality monitoring, retraining on new data, technical support for 6 months.

What's Included in the Work (Deliverables)

• Technical documentation for API and integration
• Access to dashboard and data exports
• Training session for your team (up to 2 hours)
• 6 months support and updates (SLA)

Company Metrics

Company metrics: 5+ years in AI, 30+ successful projects, 50+ clients served.

Cost and Savings

Our solution costs as low as $0.50 per response, saving up to $5,000 per conference compared to manual analysis. Typical system uptime is 98%, and we support 30+ integrated survey platforms.

Process

1. Analytics: We study your feedback streams, survey formats, past reports. We define key metrics (NPS, topic sentiment, speaker ratings).
2. Design: We choose the architecture — Hugging Face Transformers + LangChain + pgvector for semantic quote search.
3. Development: We code the pipeline (Python, PyTorch), train/fine-tune the model, configure report generation.
4. Testing: We run on historical data (if available), A/B test with manual analysis for verification.
5. Deployment: We deploy on your infrastructure (on-prem or cloud: AWS SageMaker, Google Vertex AI) or provide access to our API.

Estimated Timelines

• Basic solution (integration + standard pipeline) — from 10 business days.
• Solution with model customization and BI dashboards — up to 30 business days.

Cost is calculated individually — contact us for a free project estimate.

Experience and Guarantees

Our team has 5+ years in the AI market, with 30+ successfully implemented text analysis systems. We guarantee quality: model accuracy of at least 90% F1 on your data. We provide a certificate of model compliance with GDPR requirements (if needed).

Sample Report

A sample Executive Summary includes overall NPS, strengths, growth areas, and recommendations. The full report contains detail for each session and speaker.

Get a consultation — contact us to discuss your case. Order a pilot project — we will process data from your past event for free.

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.