Summer Project Programme · Dubai / Live Online
AI Agents for Business Applications — Summer Project Programme 2026
Bring your own project idea. Build a working AI agent and deploy it on Innosoft Gulf’s private AI infrastructure.
This hands-on programme takes you from an agent idea to a working implementation. You will use Python, local and open-source language models, approved tools, business data, and controlled workflows to develop an agent for a real professional or organisational use case.
You do not need to provide your own AI servers or assemble the entire technical platform. Innosoft Gulf provides the managed development environment, local model access and supporting services needed for the practical work. During the programme, your project can be deployed on our infrastructure.
Apply and propose your agent ideaQuick facts
From your idea to a deployed AI agent
1. Turn your idea into an agent project
Define the users, business problem, approved data sources, tools and actions your agent will support.
2. Build the agent workflow
Connect local language models to documents, vector databases, APIs, business data and approved Python tools.
3. Make it reliable and controlled
Add validation, evaluation, logging, error handling and human approval for actions that affect business systems or data.
4. Deploy and demonstrate your agent
Move beyond the notebook and deploy your working agent on Innosoft Gulf’s private cloud through Telegram, a web interface or an API.
Choose a practical AI agent project
Bring your own idea or select one of the project directions below. Each project will be scoped around a realistic workflow, approved data sources and tools, and a clear business or organisational outcome.
Lead generation and sales follow-up agent
Build an agent that works with approved prospect sources, qualifies leads, organises enquiry data, prepares personalised follow-up messages and helps track the next sales action, with human approval before communication is sent.
Real estate enquiry and property agent
Develop an agent that answers property enquiries, retrieves approved listing information, identifies prospect requirements and assists with appointments, viewings and follow-up activities.
Productivity and administrative agent
Create an agent that works with calendars, emails and business requests to schedule meetings, organise tasks, prepare responses and support approved administrative workflows.
Document processing and compliance agent
Build an agent that extracts, classifies and summarises information from tax, legal, financial, policy or contractual documents while preserving source traceability and human review.
Programme infrastructure
Develop and deploy using Innosoft Gulf’s private AI infrastructure
Participants receive controlled access to Innosoft Gulf’s private AI and data platform. The required models, databases, datasets, development environments and deployment services are provided so that participants can focus on building their agent rather than purchasing servers or configuring the complete technology stack.
Open-source AI models
Access local open-source language and vision models through Ollama. Participants can experiment with suitable models while keeping project data within Innosoft Gulf’s managed infrastructure.
Databases and project datasets
Connect agents to vector databases for semantic search and retrieval, as well as relational and other structured databases. Participants can also work with approved structured and unstructured data, including records, documents, PDFs, text, tables, images and selected large-scale datasets.
Hosted development environments
Develop the project through managed JupyterHub environments, guided notebooks and browser-based coding workspaces, with prepared Python libraries and access to the required AI, database and API services.
Kubernetes-based agent deployment
Package and deploy technically suitable agents as containerised services on Innosoft Gulf’s Kubernetes infrastructure, with access through Telegram, a web interface or an API.
Curriculum — 8 live project sessions
From language models to agentic workflows
- Chatbots, workflows and agents: practical differences
- Degrees of autonomy and appropriate human oversight
- Task decomposition, state and controlled information sources
- Running a local Small Language Model through Ollama
- Lab: build a controlled Research Agent
Reflection, critique and output improvement
- Direct generation versus review-and-revise workflows
- Generator, critic and reviser roles
- Business rubrics and measurable quality criteria
- Automated evaluation checks and human feedback
- Lab: build a reflection workflow that critiques and improves an output
Tool use, structured outputs and safe refusal
- Tool calling and function calling for business applications
- Creating approved Python tools with defined inputs and outputs
- Structured extraction and routing requests to the correct tool
- Comparing model-only answers with tool-grounded responses
- Logging tool actions and refusing when reliable evidence is unavailable
- Lab: build an agent that selects and uses approved business tools
Document intelligence and multimodal RAG
- PDF text extraction and page rendering
- Analysing tables, charts and visual page content
- Chunking, local embeddings and vector search with Weaviate
- Grounded retrieval and source-aware answers
- Local vision and language models for private document workflows
- Lab: build a multimodal document question-answering workflow
Controlled workflows and human approval
- Deterministic business rules versus model interpretation
- Read-only actions versus actions that modify business data
- Clarification, availability checks and pending actions
- Human approval before create, update or delete operations
- Calendar Agent case study: Telegram, Google Calendar and local inference
- Lab: design a multi-step workflow with an approval boundary
Reliability, validation and recovery
- Planning an action before execution
- Validating generated code or structured actions
- Repairing invalid outputs and applying restricted execution
- Database updates, verification and automatic rollback
- Visible fallbacks and audit-friendly logs
- Lab: build a reliable local customer-service workflow agent
Capstone project development
- Select and define a realistic business or organisational problem
- Design the agent architecture, approved tools and boundaries
- Combine tool use with at least two additional agentic patterns
- Add safety checks, logging and evaluation tests
- Instructor review and guided project development
Deployment, evaluation and project demonstration
- Telegram, API or web-interface integration options
- Local, cloud and hybrid deployment trade-offs
- Containerisation and Kubernetes deployment overview
- Logging, health checks, privacy and access boundaries
- Final project demonstration and evaluation
- Limitations, required human oversight and implementation roadmap
Participant feedback
What participants say about Innosoft Gulf’s AI training
Feedback from participants who attended Innosoft Gulf programmes in Agentic AI and related technical subjects.
They are truly passionate about the subject, and their energy is contagious. They create a supportive learning environment that encourages participation and critical thinking.
If you want to master AI agents, Innosoft’s AI Agents for Business Applications programme is the real deal. The progression from core basics such as RAG to complex workflows is incredibly well structured. They do not just teach theory. Innosoft provides a high-performance platform and JupyterLab environment for hands-on practice.
Great courses with plenty of real-world examples and hands-on labs. The Agentic AI training was engaging, practical and well structured. Ahmed explains challenging concepts clearly and creates a positive learning atmosphere.
It was a very practical course that provided a clear understanding of how Agentic AI works. I really appreciated the course and I am glad I attended it.
The Agentic AI course was useful. I liked the hybrid delivery because it suited my schedule. The concepts were explained well, and the individual guidance was very helpful.
Testimonials are based on participant reviews of Innosoft Gulf training programmes. Some quotations have been shortened for presentation without changing their meaning.
Prerequisites
Basic Python familiarity is recommended. Participants should be comfortable reading and modifying guided notebooks; advanced software engineering experience is not required.
- General familiarity with AI tools or language models
- Interest in documents, APIs, databases or workflow automation
- A laptop for accessing the hosted project environment
Choose your Summer 2026 cohort
Both cohorts follow the same eight-session curriculum and lead to the same programme completion standard. Choose the schedule and level of project guidance that best suit you.
For students, recent graduates and early-career participants
A group-based project cohort with guided support during the practical work and final AI agent project.
For professionals, entrepreneurs and business owners
Includes additional individual project feedback, business-use-case guidance and extended access to the development environment.
Certificate requirements
Participants who meet the following requirements receive an Innosoft Gulf certificate of completion, with KHDA attestation available:
- Attendance at the live sessions
- Completion of the guided laboratory exercises
- Submission or presentation of a final working agent prototype or structured implementation project
- Explanation of the project’s data sources, tools, workflow, safeguards and limitations
Frequently asked questions
Ready to turn your agent idea into a working prototype?
Choose the morning project cohort or the evening professional cohort, build your own agent and demonstrate it in Innosoft Gulf’s managed project environment. Places are limited to 12 participants per cohort.
Apply for the programmeQuestions? WhatsApp +971 52 351 7403 · info@innosoftgulf.com