AI-Driven Document Assistant Platform for Secure Enterprise Collaboration
Challenge
An AI software company needed to build a fully functional platform for creating and managing AI assistants powered by the OpenAI Assistants API — but their existing tools were unreliable. File uploads and retrieval behaved inconsistently depending on which OpenAI tool was in use: file search, code interpreter, and function calling each handled files differently, and the existing implementation failed to account for these distinctions, resulting in unpredictable assistant behavior and a poor user experience.
The client also needed to connect document storage from both AWS S3 and Azure Blob, enabling users to attach their own cloud-stored files to assistants — a requirement that introduced significant complexity around credential management, file routing, and per-tool handling logic.
Additionally, the platform needed to integrate with Slack so that team members could interact with assistants directly from their existing workflows — without manual onboarding steps. Users who messaged an assistant from Slack needed to be automatically provisioned, with access scoped correctly to their team’s assistants. Achieving this required reliable identity inference from Slack metadata and careful access control logic.
An AI software company needed to build a fully functional platform for creating and managing AI assistants powered by the OpenAI Assistants API — but their existing tools were unreliable. File uploads and retrieval behaved inconsistently depending on which OpenAI tool was in use: file search, code interpreter, and function calling each handled files differently, and the existing implementation failed to account for these distinctions, resulting in unpredictable assistant behavior and a poor user experience.
The client also needed to connect document storage from both AWS S3 and Azure Blob, enabling users to attach their own cloud-stored files to assistants — a requirement that introduced significant complexity around credential management, file routing, and per-tool handling logic.
Additionally, the platform needed to integrate with Slack so that team members could interact with assistants directly from their existing workflows — without manual onboarding steps. Users who messaged an assistant from Slack needed to be automatically provisioned, with access scoped correctly to their team’s assistants. Achieving this required reliable identity inference from Slack metadata and careful access control logic.
Solution
Onix built the platform as a Laravel (PHP) web application, developed on a pre-purchased template base that was significantly extended and restructured to meet the client’s requirements.
Cloud file integration was implemented for both AWS S3 and Azure Blob Storage via FlySystem, allowing users to connect their own cloud accounts and attach files to assistants. File handling logic was built specifically for each OpenAI tool type: file search uses optimized indexing to support document querying; code interpreter provides a secure execution environment for file-based calculations; function calling operates without file uploads, keeping the interface clean for that use case.
The Slack integration enables private assistant conversations directly from Slack workspaces. When a message arrives from a user with no existing account, the platform automatically creates one and infers team membership from Slack metadata — scoping access to that team’s assistants without any manual setup. A Microsoft Teams integration was scoped and partially prepared during the engagement.
Each assistant is fully configurable: users set privacy levels (private, team, or public), define OpenAI parameters such as temperature and system instructions, upload files or connect storage buckets, and add a profile image. The chat interface supports Markdown rendering, code blocks, image output, and inline file uploads. A CMS layer manages user accounts, subscription plans, shared documents, and team workspaces. The project was delivered by a team of 3 specialists in 3 months.
Onix built the platform as a Laravel (PHP) web application, developed on a pre-purchased template base that was significantly extended and restructured to meet the client’s requirements.
Cloud file integration was implemented for both AWS S3 and Azure Blob Storage via FlySystem, allowing users to connect their own cloud accounts and attach files to assistants. File handling logic was built specifically for each OpenAI tool type: file search uses optimized indexing to support document querying; code interpreter provides a secure execution environment for file-based calculations; function calling operates without file uploads, keeping the interface clean for that use case.
The Slack integration enables private assistant conversations directly from Slack workspaces. When a message arrives from a user with no existing account, the platform automatically creates one and infers team membership from Slack metadata — scoping access to that team’s assistants without any manual setup. A Microsoft Teams integration was scoped and partially prepared during the engagement.
Each assistant is fully configurable: users set privacy levels (private, team, or public), define OpenAI parameters such as temperature and system instructions, upload files or connect storage buckets, and add a profile image. The chat interface supports Markdown rendering, code blocks, image output, and inline file uploads. A CMS layer manages user accounts, subscription plans, shared documents, and team workspaces. The project was delivered by a team of 3 specialists in 3 months.
Results
The platform was delivered in full and launched as a production system within the 3-month timeline:
Complete assistant lifecycle implemented — create, configure, chat, edit, and delete, with full parameter control per assistant
Dual cloud storage integration — AWS S3 and Azure Blob connected, with per-tool file handling logic for all three OpenAI tool types
Slack integration live — real-time assistant chat from Slack with automatic user provisioning and team-scoped access control
Improved tool accuracy and assistant performance compared to the client’s previous implementation
Full CMS for subscription management, team workspaces, shared document repositories, and support workflows delivered and operational
The platform was delivered in full and launched as a production system within the 3-month timeline:
Complete assistant lifecycle implemented — create, configure, chat, edit, and delete, with full parameter control per assistant
Dual cloud storage integration — AWS S3 and Azure Blob connected, with per-tool file handling logic for all three OpenAI tool types
Slack integration live — real-time assistant chat from Slack with automatic user provisioning and team-scoped access control
Improved tool accuracy and assistant performance compared to the client’s previous implementation
Full CMS for subscription management, team workspaces, shared document repositories, and support workflows delivered and operational