Gravity Group — Mobile Applications and AI Automation

Gravity Group

⭐ 4.9 rating | 📦 24 projects | ⏱ average duration: 12 weeks

Team Composition

• Backend Developers (Java, PHP, Node.js, Bitrix, Python, 1C)
• Mobile Developers (Kotlin, Swift, Flutter, ReactNative)
• Frontend Developers (Vue, React)
• Designers
• Analysts
• Testers
• Project Managers

Technology Stack

Java, PHP, Node.js, Bitrix, Python, 1C, Kotlin, Swift, Flutter, ReactNative, Vue, React

Projects

• AI Assistant for Online Store — automation of consulting and substitute product selection
• Content Generation for Products — automatic creation of descriptions and reviews based on AI
• Project Manager Assistant — local assistant for working with documents and meetings
• AI Simulator for Sales Training — online trainer with real-time call simulation

---

Areas of Work

• 📱 Development of mobile applications of any complexity
• 🤖 Development of intelligent systems and AI-based automation

---

Case Studies

Case 1. AI Assistant for Online Store

Challenge: Products are often reserved, but this information is not updated on the website. Customers don't see real-time availability, and managers are overloaded with questions.

Solution:

• AI-powered chatbot, operating only during non-working hours for managers (weekdays from 7 AM to 4 PM Moscow time)
• Integration with inventory management system — the bot informs about real-time product availability
• Scaling: "Find Analog" button on product cards for out-of-stock items, calling the AI bot

Result:

• ✅ Reduced workload for managers
• ✅ Customers receive up-to-date information 24/7
• ✅ Potential for automating substitute product selection and consulting

---

Case 2. Content Generation for Products (Reviews + Advantages)

Challenge: Product cards lack unique texts, and there are few reviews. It's necessary to automatically generate "Advantages and features based on user reviews" blocks.

Solution:

• AI analyzes existing reviews, product descriptions, and general information from the internet
• Unique text is automatically generated, increasing conversion rates
• Fully automated process for the entire catalog

Result:

• ✅ Content uniqueness without copywriter involvement
• ✅ Increased customer trust
• ✅ Scalability to thousands of products

---

Case 3. Generation of Structured Product Descriptions

Challenge: To automatically create detailed, structured, and sales-driven descriptions for the entire catalog.

Solution: AI prepares a complete product description, including:

• Title and subtitles
• Design features
• Technical specifications
• Target audience
• Advantages
• Summary from user reviews (RU segment)

Result:

• ✅ Fully automated generation of descriptions for the entire catalog
• ✅ Savings on copywriters and editors
• ✅ Consistent style and completeness of information

---

Case 4. Automatic Selection of Recommendations and Accessories

Challenge: For each product in the catalog, related accessories need to be recommended. Manually assigning them for thousands of items is impossible.

Solution:

• AI analyzes product characteristics and automatically selects relevant accessories
• For the CHAMPION AG 252 power auger, the system recommends suitable drills, oils, fuel mixtures, and protective equipment
• Works for the entire catalog without manual intervention

Result:

• ✅ Increase in average check through cross-selling
• ✅ Time savings for managers and content managers
• ✅ Relevant recommendations for each product

---

Case 5. Project Manager Assistant

Challenge: Automate routine project manager tasks: meeting transcription, document analysis, report generation, working with the project knowledge base. Critical requirement: full locality for data confidentiality.

Solution:

Meeting Processing:

• Video transcription with speaker diarization (Whisper + pyannote) — one hour of video in 2 minutes
• Automatic participant recognition by voice
• Loading acceleration by 80–90%

Knowledge Base:

• Vector search (Qdrant + OpenSearch) — memory of all meetings and documents
• AI chat with access to all project information
• Support for any LLM (OpenAI, Anthropic, local)

Documents:

• Parsing PDF, DOCX, scans (OCR)
• Editing in OnlyOffice via browser
• AI document editing via text commands (LangGraph + Office.js)

Optimization:

• Storage compressed by 84% (65 → 10.5 MB per file)
• Client data isolation (multi-tenant)

Result:

• ✅ Local AI assistant, saving hours daily

---

Case 6. AI Simulator for Sales Training

Challenge: Create an online trainer to practice cold and warm sales skills in a safe environment, where AI plays the role of a customer with customizable behavior.

Solution:

• Real-time call simulation: AI acts as a potential client (cold/warm lead, objection level, interest)
• WebSocket architecture: real-time audio stream processing without delays
• STT/TTS: Yandex Cloud Speech Kit — speech recognition for the salesperson and voice synthesis for the AI client
• Dynamic prompt generation: adaptive context building — AI behaves naturally and unpredictably
• Gamification: salesperson's task is to convince the AI to buy

Result:

• ✅ Safe environment for sales training
• ✅ Realistic dialogues with different scenarios
• ✅ Increased sales team efficiency without risk to real clients