UNICEF works in some of the world’s toughest places, to reach the world’s most disadvantaged children. To save their lives. To defend their rights. To help them fulfill their potential.
Across 190 countries and territories, we work for every child, everywhere, every day, to build a better world for everyone.
And we never give up.
• *For every child, a digital future**
UNICEF requires an AI technical expert to develop a GenAI solution using the Microsoft Azure AI platform. The GenAI solution will enable UNICEF staff members to upload a large volume of UNICEF-internal structured and unstructured documents that are domain-specific, submit prompts and get responses as quickly and accurately as possible. UNICEF will grant access to the necessary infrastructure, regulations, and guidance to support the development and operationalization of the GenAI solution.
UNICEF developed UniBot, a GenAI/LLM that is designed to interact with specialized internal knowledge from vetted internal resources as foundational GenAI, in addition to the ability to ingest individual users' data, which is only available to that specific user to complement the foundational GenAI knowledge to produce high-quality responses. Every UNICEF staff can ask UniBot questions, however only a select group of specialists is allowed to upload additional knowledge. This is done to maintain the quality of the knowledge base and responses.
The UniBot Architecture has four high level components, an ingestion pipeline, a LLM answer flow, a front-end application, and a reporting system. Documents to be included in the knowledge base land in the Azure Storage Container, are transformed / processed by the AI Search Indexer, and placed in the AI Search Index. When a user comes on to the platform to ask a question, that question is routed through the Prompt Flow endpoint, and a response is provided back to the user. Additionally, to support transparency and maintainability, a reporting database and PowerBI reports are built off the application data. Finally, users access all the available features through the frontend application. Refer to Annex 1: UniBot diagram for additional details.
• *How can you make a difference?**
The Information and Communication Technology Division (ICTD) is currently seeking an Azure Artificial Intelligence/ Large Language Models (AI/LLM) Expert Consultant. This role will provide high level of specialization in Generative AI and will operate within the leadership of the UNICEF ICT Division.
• *Scope of Work:**
Under the supervision of the UNICEF ICT Division (ICTD) lead, the consultant will continue the development of UniBot and perform the following activities:
1. **Application Architecture Development:**
- Develop and add features to GenAI solution that leverages Microsoft Azure AI services such as Azure AI Studio, Azure Cognitive Search, Prompt Flow, etc. integrated with OpenAI’s LLMs such as GPT4 and the data sources mentioned above. Similar experience with Google Vertex or AWS Sage a plus.
- Suggest opensource options if any.
2. **Data Management:**
- Develop and maintain data pipelines for handling a high volume of structured and unstructured documents uploaded by a user, such as:
- Upload files and folders
- SharePoint integration and content transformation
- Web scraper
- Ensure data security and compliance with relevant UNICEF data protection regulations.
3. **User Administration**
- Implement roles and users’ groups using Microsoft Entra ID
- Create basic UI for users’ administration.
4. **User Interface Development:**
- Build a simple, user-friendly UI/UX for UNICEF staff members to upload documents and interact with the GenAI.
- Implement features to automate document upload and chat-based querying.
5. **Cost** **Tracking**
- Implement a feature to track and display the cost associated with chat interactions.
- Implement a feature to track backend transactions.
- Implement basic reports with options for duration, frontend and/or backend transactions.
6. **L****ogs,** **Usage Reports and History**
- Log timestamped training events, algorithm change events, prompts, responses, user assessment (thumbs up/down, hallucination, bias, accuracy)
- Implement Provide method(s) to manage log size, backups and archival.
- Create basic usage reports and guidance for PowerBI users.
- Provide users with access to chat history for reference and review.
- Provide the ability to reset the session/start new session.
7. **Prompt Engineering and Model Fine tuning:**
- Document efficient prompt examples that help the LLMs generate accurate and relevant responses.
- Document the best ways to train and fine-tune LLMs for domain-specific knowledge and applications.
- Employ RAG, Prompt or other techniques to anchor the GenAI responses to the provided data sources and establish techniques to mitigate hallucinations and increase accuracy.
8. **Security and Compliance:**
- Ensure compliance with UNICEF security, ethics, and pr
Location: Anywhere
Posted: Oct. 23, 2024, 8:57 a.m.
Apply Now Company Website