Bladex AI

Main Dashboard (all your chats, advanced Search Bar which also supports Query Input all the feature can accesed by that Advance search bar)
you can enable light mode from search bar just toggle it
AI agents creation section
connect your platform

Inspiration

It started with a simple observation when i was working as an Intern . Every morning, HR faces 200+ messages: "Any waiter positions in Mumbai?" "Experienced chef looking for work" "Need hotel receptionist job urgently"

Her team wastes hours copy-pasting responses, asking for CVs, while qualified candidates slip away.

every small business owner, founder, and support team is drowning. They're juggling DMs on Telegram, Whatsapp responding to support tickets, and trying to remember if "John D." from last week is the same "John" asking for a refund today.

We saw that system was chaotic and then we had been on this a week ago but coding started yesterday and with cursor it became way faster. So to solve that problem first we bought domain and then started working.

We envisioned an integrated, intelligent platform that doesn’t just answer questions, but acts as a true co-pilot for business communications. A system that remembers, understands, and acts on your behalf. This is Bladex AI.

What it does

Bladex AI is an all-in-one, AI-native platform designed to unify and automate customer conversations. It’s not just a chatbot; it's an operating system with a brain, giving businesses an autonomous agent to manage their entire communication workflow. Bladex AI brings to the table:

Unified Command Center : Connect all your channels (starting with Telegram) into a single, elegant dashboard. No more tab-switching chaos.

AI Agents: Our sophisticated AI Orchestrator analyzes every incoming message and routes it to the perfect agent—a lightning-fast Rules-Agent for common questions, a deep-thinking LLM-Agent for complex queries, or a Classifier-Agent to understand intent.

Long-Term Memory: Powered by vector embeddings, Bladex AI remembers crucial details across conversations. It knows who your customers are, what they’ve asked for before, and uses that context to provide deeply personalized interactions.

Natural Language as Your CLI: Forget clunky interfaces. Our QueryBox lets you command the entire platform with plain English. Type "Show me all messages from Manish about 'Project X' last week," or "Reply to the contact who mentioned 'Chef Mumbai' and tell them congratulations!"—and watch it happen.

A revolutionary multi-tenant architecture gives every single user their own dedicated, isolated database. Your data is yours, period.

How we built it

We made Bladex AI from the ground up for security, intelligence, and scale.

The Brain - AI Orchestration System: At the core is our orchestrator (orchestrator.ts). When a message arrives, it's not sent to a single model. Instead, multiple specialized agents (built with Anthropic Claude) bid to handle it. The orchestrator scores their confidence and deploys the best one for the job, ensuring optimal speed and accuracy.

Multi-Tenant Architecture: We added two distinct database layers using Supabase: Admin Database: (this is for developer who will keep testing new features and ship all they have to do is login with /admin page and this uses admin supabase db.)

User Databases: As we mentioned before we provide isolated database for each user mean chat history will be saved in their isolated db, data is not leaking anywhere.

It was one of toughest things to implement but we did we just implemented connect with supabase feature when user click on that button then all the credentials will be directcly featched and save them into admin db and cookies and now whenever db is called it will select only user db.

The Memory Vector Embeddings: We use pgvector in Supabase and Gemini's embedding models to give our AI a persistent memory. Key information is extracted from conversations, converted into vectors, and stored, ready for semantic retrieval to provide rich context for any future interaction.

Tech Stack: Framework: Next.js 14 (with Server Components) Styling: Tailwind CSS Database & Auth: Supabase (with pgvector) AI Models: Anthropic Claude Integrations: Telegram Bot API Deployment: Vercel

Challenges we ran into

The Multi-Tenant Webhook Problem: How does our system know which user's database to write to when a webhook from Telegram arrives stateless, with no browser context? We solved this by creating a central bot registry in the admin database that maps bot tokens to encrypted user database credentials, allowing for secure and instant routing. Designing the AI Orchestrator: Building a system where AI agents compete and collaborate was complex. setting up scoring logic to choose the right agent (fast vs. smart vs. rule-based) for the right task was a significant challenge that required extensive prompt engineering and logical design.

Credential Security: Managing credentials for dozens of dynamic databases demanded a robust security model. We implemented a server-side encryption/decryption layer to ensure user database keys are never exposed and are only used ephemerally to establish a connection.

Accomplishments that we're proud of

We didn't just build a feature; we built a new architectural paradigm for AI-SaaS. We are incredibly proud of:

Successfully implementing a true multi-tenant database architecture—an enterprise-grade feature—in a hackathon timeframe. Creating an intelligent AI Orchestration system that goes beyond simple LLM calls, demonstrating a more mature, agentic approach to AI. Building the QueryBox, a feature that feels like magic and showcases the true power of function-calling and natural language interfaces.

What we learned

We learned that the next generation of AI applications won't be simple wrappers around an API. We need proper flow with traditional software engineering LLM is just an API all the logic would be written in coding always so it is important to learn the base fundamental and how we can optimize any system.

What's next for Bladex AI

Our vision is to make Bladex AI the central system for all business communications.

More Channels: Integrate WhatsApp and Email. Custom Agent Builder: A simple UI for users to write their own specialized agents in simle natural language.