Projects

Context-Aware Chatbot with Role-Based Access

April 2022
I am the creator of a context-aware and role-based access chatbot, a cutting-edge Retrieval Augmented Generation chatbot meticulously engineered to elevate user interactions by delivering succinct and accurate responses tailored to the provided context. The chatbot adheres to a robust set of guiding principles, ensuring optimal performance and user satisfaction:
Contextual Precision: The chatbot relies exclusively on the contextual cues at hand to craft responses, guaranteeing relevance and precision.
Acknowledging Gaps: In cases where the context is insufficient, The chatbot transparently acknowledges its limitations, fostering trust and openness in interactions.
Code Clarity: When presenting code examples, The chatbot employs triple backticks for proper formatting, specifying the programming language to enhance clarity and understanding.
Focused Information: The chatbot avoids generating pseudo-code or unrelated information, maintaining consistency by delivering content that is consistently on point.
Conciseness: Prioritizing brevity and clarity, The chatbot encapsulates responses within 600 characters or less, ensuring efficient and effective information delivery.
Beyond these core features, The chatbot is designed with advanced capabilities:
Role-Based Access Control: The chatbot is aware of the user's role, distinguishing between trainees and administrators. Trainees can access public data/information, such as details on 10 academy. Administrators, on the other hand, have privileged access to confidential or private information, ensuring the right level of data security.
Conversation Chain Retention: The chatbot possesses the unique ability to retain and comprehend past conversation chains. This feature enables a seamless and coherent flow of interactions, making each conversation feel like a continuation rather than a series of disjointed queries.
Feedback-Driven Learning: The chatbot is continuously evolving through user feedback. Each interaction serves as an opportunity for learning and adjustment, allowing the chatbot to adapt and enhance its responses over time. This iterative learning process ensures a personalized and improving conversational experience for users.
The chatbot represents a leap forward in chatbot technology, combining contextual awareness, role-based access control, and conversation chain retention to provide users with an unparalleled and tailored conversational experience.
Used techstacks: Chatbot, Prompt Engineering, Python, Vector Database, LLM Prompt, AI Bot, AI Chatbot, Generative AI, Machine learning

Speech data collection using kafka cluster

April 2022
Our project centers on leveraging Apache Kafka, an open-source distributed event streaming platform, to construct an advanced data pipeline. This pipeline integrates messaging, storage, and stream processing capabilities, facilitating the collection, storage, and real-time analysis of audio datasets. Key components include data producers and consumers, an Apache Kafka cluster, connectors for seamless integration with Amazon S3 buckets, and an Apache Spark Stream preprocessor. This setup ensures efficient data flow from data sources to storage, enabling robust analytics and insights generation from large-scale speech datasets. By harnessing Kafka's scalability and real-time capabilities, we aim to enhance data processing efficiency and enable rapid insights extraction in the domain of speech data analytics.
Used techstacks: AWS S3 Bucket, AWS athena, spark, kafka, ELT pipeline, Airflow, docker

GitHub Link

Data warehouse for pNEUMA | open-traffic datasets

June 2022
In June 2022, we undertook the development of a robust and scalable data warehouse solution tailored for handling open-traffic datasets within the framework of pNEUMA. This initiative involved the implementation of a comprehensive ELT (Extract, Load, Transform) pipeline leveraging MYSQL Database as the foundational storage backend. The pipeline was augmented with dbt (Data Build Tool) for advanced data transformations, ensuring data quality and consistency across the warehouse.
Airflow, a powerful orchestration tool, was integrated to automate and schedule workflows, managing the seamless execution of data ingestion, transformation, and loading processes. This orchestration capability not only enhanced operational efficiency but also facilitated the timely delivery of updated insights and analytics-ready data to stakeholders.
Redash, a dynamic visualization platform, was deployed to provide intuitive and interactive data visualizations and dashboards. This visualization layer empowered users to explore and derive actionable insights from the aggregated and transformed data housed within the warehouse.
Overall, our approach aimed to optimize data handling, ensure data reliability, and empower data-driven decision-making within the domain of open-traffic datasets, supporting pNEUMA's mission with robust analytics capabilities.
Used techstacks: dbt, ELT pipeline, Airflow, Redash.

GitHub Link

USGS LIDAR Data warehouse

Sep 2022
Our focus was on understanding water flow dynamics within maize farm fields, crucial for advancing research on new agricultural products. Maize yields vary significantly across fields despite uniform farming practices, necessitating a deeper understanding of spatial variability. To enhance research accuracy, we developed a user-friendly Python module. This module empowers domain experts and data scientists to seamlessly access, visualize, and transform satellite and LIDAR data. Specifically, it interfaces with the USGS 3DEP API to fetch high-resolution elevation data, enabling precise analysis of field characteristics. Our goal was to provide reliable insights into maize productivity variations, ensuring that experimental changes like new fertilizers yield meaningful outcomes, free from random environmental influences.
Used techstacks: LASzip, PDAL, CloudCompare, USGS 3DEP API and GIS software packages such as QGIS or ArcGIS.

GitHub Link

Speech Recognition: Amharic and Swahili Languages

March 2022
This project focuses on developing a robust speech recognition system capable of accurately converting spoken Amharic and Swahili languages into text. The goal is to facilitate seamless communication and interaction in regions where these languages are predominant, enhancing accessibility to digital tools and services. Leveraging machine learning frameworks like TensorFlow and PyTorch, the model integrates advanced signal processing techniques such as Fourier transforms, short-time Fourier transforms, and Mel-frequency cepstral coefficients (MFCCs). These methods are pivotal in capturing and analyzing speech patterns, ensuring high accuracy in transcription and improving usability across diverse linguistic contexts.
Used ML frameworks like TensorFlow, PyTorch and NLP
Used techniques like Fourier transforms, short-time Fourier transforms and Mel-frequency cepstral coefficients (MFCCs).