Software engineer with over 15y of experience in various fields, including gamedev, web development and cyber security.

github

projects

devto

hashnode

substack

Elijah Rastorguev • Peerlist

<h3>What TozheBot Is</h3>TozheBot is a platform for creating and managing AI agents that help automate user communication and everyday business processes.With TozheBot, you can create AI agents with custom instructions, connect knowledge bases, configure tools, and integrate agents into your existing communication channels.Agents can answer user questions, help people find information, assist customers, handle support requests, and perform other tasks within their configured capabilities and permissions.Instead of developing a separate bot for every task, you can use a single platform to create and manage multiple AI agents.<h3>What Can You Do with TozheBot?</h3>Automate Customer SupportCreate AI agents that answer frequently asked questions, use information from your documentation, and help users resolve common issues.Work with Knowledge BasesConnect websites, documents, and other information sources so that agents can generate responses based on your company's materials.Communicate with Users Across Multiple ChannelsUse available integrations to interact with users through your website and supported messaging platforms.Automate Sales and Lead ManagementConfigure agents to provide initial consultations, qualify inquiries, collect the necessary information, and forward potential customers to your sales team.Create Specialized AI EmployeesConfigure individual agents for different tasks, define their instructions, choose their available tools, and organize their work around your team's processes.<h3>How TozheBot Works</h3>The platform is built around several core components.AI Agent — an entity that interacts with users and performs assigned tasks according to its configured instructions.Instructions — rules that define the agent's role, communication style, workflow, and limitations.Knowledge Base — a source of information that an agent uses to prepare responses.Tools and Integrations — additional capabilities that allow agents to interact with connected services and perform supported actions.Communication Channels — the points of interaction between agents and users, such as a website widget or a supported messaging platform.Analytics and Management — tools for monitoring agent performance, evaluating results, and adjusting agent behavior.The availability of individual features depends on your configuration, connected integrations, and subscription plan.<h3>Who Is TozheBot For?</h3>The platform can be used in a variety of scenarios:<ul><li>Customer Support: Answer common questions and help users get the most out of your product.</li><li>SaaS and IT Products: Provide assistance with product documentation, features, and troubleshooting.</li><li>E-commerce: Answer product questions and help customers choose the right items.</li><li>Gaming Projects: Assist community members and answer questions about game rules, mechanics, and documentation.</li><li>Sales and Marketing: Handle initial inquiries and qualify potential customers.</li><li>Internal Operations: Help employees find information and work with company documentation.</li></ul>

Tozhebot

Open Guard Layer (OGL-Mini) - a full-featured, fast and lightweight security model for AI agents with ready-to-use TypeScript, Python and Go modules.<h3>❓ Why OGL-Mini?</h3>🔹 Ready-to-Use Lightweight modules for TypeScript / Go / Python; 🔹 Extra lightweight model (~300MB). Inference starts at CPU; 🔹 Extremely fast (great for realtime applications); 🔹 Production ready with benchmarks; 🔹 Russian / English languages as primary;<h3>🔒 What kind of attacks does it protect against?</h3><ul><li>General Attacks: LLM01 Prompt Injection + LLM02 Sensitive Disclosure, LLM03 Supply Chain, LLM04 Data Poisoning, LLM05 Improper Output, LLM06 Excessive Agency, LLM07 Misinformation, LLM08 Hidden Context, LLM09 Vector Weakness, LLM10 Unbounded Consumption + Agentic (Goal Hijack, Privilege Abuse, Code Exec, InterAgent, Tool Misuse, Memory Poisoning, Cascading, Rogue, Policy Puppetry, EchoLeak, Lies-in-the-Loop)</li><li>Modern Injections: S3 encoding (base64), homoglyph, zero-width (\u200b), spaced letters, control tokens (&lt;|im_start|&gt;, [INST]), indirect JSON/tool, HTML markdown, agent-specific, best-of-n, typoglycemia - all in training (14k modern obfuscation, bilingual)</li><li>Auto-sanitizing: Sanitize fields by length, bidi-characters, control characters, excessive whitespaces etc.</li></ul><h3>Usage</h3><ul><li>✅ Input Guard - block prompt injection before agent (&lt;10ms p95, RU/EN, modern obfuscation)</li><li>✅ Output Guard - block PII / system-prompt leaks</li><li>✅ PII detection &amp; redaction - 13 regex + 11 ONNX types (PERSON, EMAIL, PHONE, IP, IBAN, BANK_CARD, PASSPORT, GOV_ID, DOB, ADDRESS, SOCIAL, MAC)</li></ul><h3>Our model is</h3><ul><li>Polyglot architecture for microservices. We support Python / TypeScript / Go implementations.</li><li>Built-in lightweight data leak prevention (PII NER): For AI agents that process personal data, this is a critical security feature.</li><li>Unique lightweight design and WASM: This allows security to be pushed to the perimeter. For example, to the user's browser or IoT device - a feature unavailable with other solutions.</li></ul>

Open Guard Layer (OGL-Mini)

Virton AI is a next-generation AI-powered virtual fitting room for online stores.Who it's for:This product is designed primarily for online store owners and their customers. It helps them make purchasing decisions, while simplifying their experience with product cards.Problem:The main problem is the long decision-making cycle before purchasing. To understand whether clothing will match their look, customers must either visit a showroom or wait for delivery.Solution:We offer a shortened decision-making cycle through virtual AI try-ons. Users upload photos, select items, and can immediately determine how well a particular look will suit them.Advantages:<ul><li>We don't charge for each generation;</li><li>AI models are individually trained for each customer and their product range;</li><li>Our algorithm preserves the user's context and image as much as possible;</li></ul>Coming soon:<ul><li>Video fitting (up to 1 minute);</li><li>Select clothing sizes without leaving the online store using your photos (up to 4 photos for size accuracy), not just a visual fitting.</li><li>Assistant for choosing clothes in online stores (virtual stylist).</li></ul><a target="_blank" rel="noopener noreferrer" class="break-all link-break text-gray-500 hover:underline" href="https://viron.tech/">Try it for free.</a>