
Average AI response time — instant answers replacing hours/days of waiting
Specialist agent domains covering every business function from products to procurement
Reduction in manual inquiry triage through automated intent classification and routing
Feroze1888 relied on separate static web forms for Marketing, HR, Procurement, and General inquiries — a fragmented system that created response delays ranging from hours to days. Global partners, job seekers, investors, and suppliers in different time zones had no way to get instant answers about products, open bids, financial data, or career opportunities. The support team was overwhelmed with repetitive queries, while complex domain-specific questions (e.g., investor governance details, sustainability metrics, or supplier registration steps) frequently got misrouted, requiring multiple back-and-forth emails to resolve.
We built a FastAPI-based orchestration layer that connects WhatsApp (via Evolution API) to a network of 8 specialist AI agents, each trained on a curated knowledge base scraped directly from Feroze1888's official website. A Router Agent uses keyword fast-pathing augmented by LLM-based intent classification to dispatch messages to the correct specialist in under 500ms — Product, Company, Career, Investor, Sustainability, Procurement, General, or Escalation.
A Router Agent classifies incoming WhatsApp messages into 8 intent categories using a hybrid fast-path keyword matcher and LLM-based fallback, achieving sub-second dispatch to the correct specialist domain.
Dedicated AI agents for Products, Company Info, Careers, Investor Relations, Sustainability, Procurement, General Support, and Human Escalation — each trained on a structured, domain-specific knowledge base with confidence self-assessment.
Multi-layer defense system including rate limiting, prompt injection detection, off-topic blocking, PII redaction, output sanitization, and escalation detection — ensuring safe, on-brand responses at all times.
In-memory per-session context stores the last 3 exchanges per user, enabling natural follow-up conversations without repeating context — with automatic escalation when user frustration is detected.
Real-time typing indicators ('composing...', 'paused') for human-like interaction, auto-splitting of long messages with progress markers, and WhatsApp-native bold/emoji formatting across all agent responses.
Full microservices stack (FastAPI backend, Evolution API gateway, PostgreSQL, Redis) deployable via `docker-compose up -d` with Nginx reverse proxy support, health checks, and isolated networking.