Page 52 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

Justify UN Goals selection:

1. SDG 9 (Industry, Innovation, and Infrastructure): The AI-based system supports building
resilient infrastructure by ensuring integrity in the procurement process, which is crucial
for developing quality, reliable, and sustainable infrastructure.
2. SDG 11 (Sustainable Cities and Communities): Enhancing the tendering process
contributes to the development of sustainable cities through accountable and transparent
governance, which can lead to the efficient creation of public services and infrastructure.
3. SDG 16 (Peace, Justice, and Strong Institutions): By combating corruption and enhancing
transparency, the AI initiative contributes significantly to building effective, accountable,
and inclusive institutions at all levels, which is at the core of Goal 16.

AI aids in achieving these SDGs by providing a sophisticated tool to analyze vast amounts
of tender-related data, identify patterns indicative of corruption, and provide actionable
intelligence. This capability ensures that resources are allocated efficiently and fairly, thus
fostering innovation (SDG 9) and aiding in developing sustainable infrastructure and
communities (SDG 11). Moreover, promoting a corruption-free environment reinforces
the institutions' integrity and credibility, which are crucial for maintaining peace and justice
(SDG 16). AI speeds up the detection process and provides a non-biased system that can be
continuously improved, scaling up the impact of anti-corruption efforts to drive sustainable
development in Tanzania.

Partner name: University of Dar es Salaam

Partner name: University of Dodoma

8�2�2 Future work

Standards development related to the use case, Others Elaborate proposal: This project aims to
develop an AI-powered system to combat corruption and promote transparency in Tanzania's
public procurement process and project implementation cycle. The proposed methodology
outlines a comprehensive approach leveraging state-of-the-art natural language processing
(NLP) and large language model (LLM) techniques.

The core components of this use case include:
1. Data Collection: Gather a comprehensive dataset from various sources, including the
Prevention and Combating of Corruption Bureau (PCCB), district councils, Tanzania
courts, tax authorities, and public procurement regulatory bodies. The dataset will
encompass tender documents, evaluation reports, corruption cases, laws, and regulations
related to procurement.
2. Unsupervised Training and Data Preprocessing: Utilize NLP techniques and LLMs to
preprocess and analyze the collected data. This includes the preprocessing stage, which
removes duplicate text and irrelevant data elements, as well as personal or organizational
information, and then unsupervised pre-training (foundation model development) of
LLMs on the dataset to gain a general understanding of the procurement domain.
3. Supervised Fine-tuning: Fine-tune the pre-trained LLM using supervised learning
techniques for specific tasks such as sentiment analysis, named entity recognition, text
classification, anomaly detection in tender documents, and text generation. The goal
is to develop models capable of different downstream tasks, such as analyzing tender
documents, identifying irregularities, providing a corruption probability percentage, and
text generation.
4. Document Analysis and Corruption Detection: Deploy the fine-tuned model to analyze
tender documents and identify potential fraud, corruption risks, and irregularities. The

36