AI agent platform for modern businesses
One Stop Agent (OSA) was built as a SaaS platform for businesses to create, train and deploy AI agents using their own knowledge. The challenge was to turn fragmented content across PDFs, websites and internal documentation into reliable conversational agents, while handling the infrastructure behind retrieval, real-time conversations, human escalation, security and usage at SaaS scale.
Enterprise AI · SaaS
2025–2026
AI engineering · Platform architecture · Product engineering
MVP to production SaaS platform
Approach
We designed OSA around a simple principle: businesses should be able to deploy useful AI agents without having to build and operate the underlying AI infrastructure themselves.
Each agent combines business-specific knowledge with configurable instructions and conversational context. A retrieval pipeline transforms documents, websites and text into searchable knowledge, allowing agents to ground responses in relevant enterprise information rather than relying solely on the underlying model.
The architecture also treats AI as one part of the support workflow. Conversations can move from autonomous AI responses to human support in real time, while retaining conversation context and history.
Engineered
A multi-tenant AI agent platform covering the complete lifecycle from agent creation and knowledge ingestion through retrieval, inference, deployment and monitoring.
The knowledge pipeline ingests PDFs, raw text and crawled website content, stores source files in AWS S3, parses and chunks content, generates 1,536-dimensional embeddings and indexes them in Pinecone using cosine similarity. At inference time, semantic retrieval identifies relevant context, which is combined with system instructions and conversation history before being passed through LangChain to GPT-4o.
A real-time conversation layer built with Socket.io streams agent responses and powers human handoff. Conversations can transition into a waiting state, notify available support users and continue as live human conversations without creating a separate support channel.
The platform includes embeddable JavaScript chat widgets, multi-tenant agent administration, usage metering, Stripe subscriptions and webhook-driven billing, multilingual interfaces, MFA and operational monitoring. TAP integration adds an agent identity and verification layer, including registration, verification state management and decentralized identifiers for verified agents.
- Next.js
- ·React
- ·TypeScript
- ·Node.js
- ·Express
- ·LangChain
- ·OpenAI GPT-4o
- ·Pinecone
- ·Prisma
- ·MySQL
- ·Socket.io
- ·AWS S3
- ·AWS SES
- ·Stripe
- ·Sentry
- ·JWT
- ·TOTP MFA
Outcomes
Knowledge to agent
Converted PDFs, websites and business content into deployable AI agents through an automated ingestion, embedding and retrieval pipeline.
AI + human support
Combined 24/7 AI assistance with real-time escalation to human operators when conversations required intervention.
Production-ready agent infrastructure
Built the supporting SaaS capabilities around AI including authentication, MFA, billing, usage controls, observability, multilingual support and embeddable deployment.
Verifiable agent identity
Integrated TAP verification to associate agents with identity, ownership and decentralized identifiers, creating a foundation for more trusted agent interactions.
Built as the technology platform behind One Stop Agent, covering the full lifecycle from enterprise knowledge ingestion to AI inference, deployment, human handoff and agent verification.