Hukum.tech

Self-Service AI Chatbot Platform & Automated Customer Support Ecosystem

🤖 LLM Chatbots ⚡ Flask (Python REST API) ⚛️ React (Vite) 💾 Supabase (PostgreSQL RLS) 📱 WhatsApp Cloud API ☁️ DigitalOcean
🚀

Live Commercial SaaS Platform

Automates repetitive customer support inquiries across web widgets and WhatsApp with real-time prompt tuning and human operator fallback.

Visit hukum.tech ↗

Executive Summary

Hukum.tech is a proprietary, multi-tenant SaaS platform built to solve the high costs, human latency, and operational friction of modern customer support. The system empowers small and medium-sized businesses (SMBs) to deploy custom context-aware AI chatbots across both their web storefronts and official WhatsApp Business accounts without writing code.

The Problem: Operational Overhead in Support Channels

Modern online businesses face continuous customer inquiries regarding order status, shipping policies, product catalogs, pricing tiers, and booking reservations. Handling these inquiries manually introduces significant drawbacks:

  • High Payroll & Slow Turnarounds: Support representatives cannot provide 24/7 sub-minute response times without multi-shift hiring.
  • Fragmented Channels: Web visitors and WhatsApp users receive inconsistent information when handled by different team members.
  • Rigid Rule-Based Bots: Traditional decision-tree chatbots fail on conversational nuances, misspelling, and compound questions.

The Solution & Technical Architecture

To resolve these challenges, I architected a unified AI chatbot infrastructure capable of answering customer questions with semantic accuracy, parsing unstructured business knowledge on the fly, and routing complex queries to human operators when needed.

[Client Channels] Web Widget (JS SDK) & WhatsApp Cloud API
    ↓ Webhooks / HTTPS REST
[API Gateway & Router] Python Flask Backend on DigitalOcean (Rate Limiter, Token Buffer)
    ↓ Dynamic Prompt Compilation
[Intelligence Engine] LLM API Matrix (Context Window, Guardrails, Persona Rules)
    ↓ Session State & RLS Policies
[Database & Real-Time] Supabase PostgreSQL (Tenants, Conversation Logs, Handover Triggers)
    ↓ Live Dashboard Sync
[Operator Portal] React (Vite) Admin & Live Chat Interceptor

My Role & Key Responsibilities

As the Full-Stack & AI Automation Engineer, I designed and executed the entire product lifecycle from ground up:

  • Backend API Engineering: Built the modular Python Flask REST API responsible for webhook ingestion, prompt composition, token budgeting, and third-party communication.
  • Database Design & Security: Architected the relational schema in PostgreSQL via Supabase, creating automated triggers, audit logs, and rigorous Row-Level Security (RLS) policies to ensure strict tenant data isolation.
  • Frontend Dashboard: Developed the client administration console in React (Vite), providing live prompt tuning, conversation analytics, and operator intervention controls.
  • Multi-Channel Integrations: Configured Meta WhatsApp Business Cloud API webhooks with HMAC validation, retry queues, and mobile payload rendering.

AI & Automation Innovations

  • Self-Service Prompt Matrix: Business owners can alter bot personality, conversational tone, and strict domain boundaries in real time from the dashboard. Changes take effect instantly without restarting services or redeploying code.
  • Dynamic FAQ Document Ingestion: Allows users to upload PDF or DOCX policy sheets. The backend parses unstructured text blocks, verifies integrity, and embeds the sanitized rules straight into the bot's runtime prompt context.
  • Human-in-the-Loop Handover: When sentiment drops or a customer requests human assistance, the engine transitions the conversation to "human-operator" state, muting automated AI responses and alerting human agents inside the React console.

Technical Challenges & Resolutions

Challenge 1: Webhook Concurrency & WhatsApp Retries

Meta WhatsApp Cloud API retries unacknowledged webhooks aggressively within 5 seconds. Long LLM generation times previously caused duplicate messages. I implemented asynchronous webhook acknowledgment (HTTP 200 returned in <100ms) with an internal Redis/memory buffer to process generations independently.

Challenge 2: Multi-Tenant Data Isolation

Preventing cross-tenant data leakage in multi-tenant SaaS environments requires foolproof safeguards. I wrote native PostgreSQL Row-Level Security (RLS) policies bound to authenticated user JWTs, ensuring tenant queries are isolated at the database engine level.

Production Impact & SaaS Tiers

In our tested workflow, Hukum AI resolved up to 80% of supported customer inquiries without human intervention, significantly reducing repetitive triage for business operators while routing complex requests seamlessly to live staff.

Hukum.tech operates with three clearly defined commercial tiers:

Basic Tier

Rs 5,000 / mo
Web widget chatbot, 1 managed business entity, 10,000 monthly messages, self-service live prompts, and standard operator support.

Pro Tier

Rs 10,000 / mo
Dedicated WhatsApp bot, 30,000 monthly messages, custom tone adjustments, FAQ document parser, and standard support channels.

Enterprise Tier

Rs 20,000 / mo
Web + WhatsApp bot channels, unlimited messages, priority support, dedicated WABA client setup, and custom third-party integrations.

Learn more about how I build conversational AI solutions on the AI Automation services page or read the technical deep-dive: How I Built an AI Customer Support Agent with n8n & Webhooks.