AVYANTA ElectroShield

The product website for AVYANTA, an Indian electrical safety technology company. Its flagship product, ElectroShield, is a helmet-mounted device that detects high voltage without contact and warns the wearer before they get close enough to be hurt.

Two things. It argues with data — a decade of national electrocution statistics and an interactive state map, so the risk is undeniable before the product is even mentioned. And it ships with a custom AI assistant that can answer technical questions properly.

Retrieval-augmented generation. Instead of relying on what a language model happens to know, the assistant first searches the company's own documents for relevant passages, then answers using them. For a safety device with specific voltage ratings, that difference matters.

A Python backend on FastAPI with LangGraph orchestrating the agent, ChromaDB as the vector store and sentence-transformers generating embeddings locally. The language model is Llama, served through a fast inference API.

Yes — that's what the intent routing is for. Every message is classified first, and when someone shares an email or asks to be sent pricing or documentation, it's handled as an action rather than a chat reply, so the enquiry reaches the team.

Layered defences: pattern screening for known jailbreak phrasings like “ignore previous instructions”, an LLM-based safety assessment, and a retrieval step that anchors answers to approved documents. The assistant stays on its subject.

Yes — it holds conversational state, so follow-up questions work naturally. You can ask about a specification and then say “and what about in rain?” without repeating the context.

Yes — this is exactly what our AI solutions work covers. We ingest your manuals, specifications or policies into a vector store and build an assistant that answers from them, with guardrails suited to your industry.

React with Vite and Tailwind CSS, using Framer Motion for the scroll and counter animations and Lenis for smooth scrolling. Thirteen pages and around seventy-four components in total.

Yes — data-led product and B2B websites are a strong fit for us, with or without an AI layer. If you have a technical product that needs explaining before it can be sold, that's the kind of brief this was.