Isaac Kobby Anni | AI Researcher | SLM-First Agentic Systems

About Me

Isaac Kobby Anni

AI Researcher & PhD Candidate

I'm a PhD candidate in Data Science at Bowling Green State University, researching SLM-First Agentic Systems—next-generation AI architectures built on Small Language Models (SLMs) that prioritize privacy, cost-efficiency, and controller-mediated orchestration. My work focuses on developing agentic AI systems that achieve accuracy parity with large language models while reducing costs by 10-100× and enabling on-premises deployment.

My dissertation explores three interconnected research thrusts: specialized feature engineering for SLMs, case-based reasoning for agentic workflows, and multi-agent AutoML systems. I'm particularly interested in the intersection of AI efficiency, privacy-preserving architectures, and practical deployment challenges.

Agentic AI & SLMs

Small Language Models, Agentic Systems, Controller-mediated architectures, Multi-agent orchestration

Privacy & Efficiency

Privacy-preserving AI, On-premises deployment, Cost optimization, Latency reduction

Research Focus

Feature engineering for SLMs, Case-based reasoning, Multi-agent AutoML, Model efficiency

Technical Stack

Python, PyTorch, Transformers, LLM frameworks, Distributed systems, MLOps

Email Me Download CV

Skills & Expertise

Technical foundations for agentic AI research and production systems

Agentic AI

Multi-agent orchestration, tool use, controller-mediated workflows

Small Language Models

Fine-tuning, efficient inference, SLM-first system design

NLP & ML

Transformers, representation learning, text classification, RAG

Privacy & Deployment

On-premises AI, cost optimization, safety validation

Python & PyTorch

Research prototyping, model training, evaluation pipelines

Computer Vision

Classification, semantic segmentation, multimodal learning

Predictive Analytics

Forecasting, feature engineering, AutoML workflows

Data Engineering

ETL pipelines, SQL, analytics dashboards (Power BI)

Research

SLM-First Agentic Systems: Privacy-preserving, cost-efficient AI architectures

Dissertation Research Overview

My PhD research focuses on developing SLM-First Agentic Systems—next-generation AI architectures that leverage Small Language Models (SLMs) to achieve accuracy parity with large language models while dramatically reducing costs, latency, and privacy risks.

10-100× Cost Reduction

Achieving significant cost savings compared to LLM-based systems

Privacy-Preserving

On-premises deployment enabling data sovereignty

Sub-300ms Latency

Real-time performance for production workloads

Research Thrusts

Project 1

Feature Engineering

Controller-Mediated Feature Engineering

Developing specialized feature engineering controllers for SLMs that handle temporal, categorical, and numerical data types with model sizes ranging from 1.5B to 2.4B parameters.

Controller: Orchestrates feature processing
Models: Temporal (2B), Categorical (2B), Numerical (2.4B)
Escalation: 7B/LLM fallback for complex cases
Targets: 10-30× cost reduction, sub-300ms latency, on-premises deployment

Project 2

Case-Based Reasoning

Agentic Case-Based Reasoning

Building iterative case-based reasoning systems with specialized agents for case retrieval (1.5B), revision (3.8B), and one-pass deployment scenarios.

Development: Case Bank, Retriever (1.5B), Revisor (3.8B), Iterative refinement
Deployment: One-pass (3.8B), Controller, Auto-validation
Performance: 10-100× cost reduction, 85%+ one-pass success rate
Privacy: On-premises deployment with data sovereignty

Project 3

Multi-Agent AutoML

Orchestrated Multi-Agent Systems

Developing a planner agent (3.8B) that orchestrates specialized agents for data preparation, feature engineering, model selection, hyperparameter tuning, and evaluation.

Planner: 3.8B orchestrator agent
Specialized Agents: Data (1.5B), Feature (2B), Model (3.8B), Hyperparameter (2B), Evaluation (1.5B)
Efficiency: 70-90% context reduction, sub-5 minute execution
Parallelization: Concurrent agent execution

Evaluation Framework & Success Criteria

Core Metrics

Accuracy: Parity with LLM-based baselines
Cost: ≥10× reduction compared to LLM APIs
Latency: Sub-300ms for real-time applications
Energy: Reduced computational requirements
Privacy & Safety: On-premises deployment, programmatic verification, schema validation

Cross-Cutting Methods

Training: Wind tunnel testing, WSD scheduling, hyperparameter transfer, continuous training
Safety: KG-backed checks, privacy-preserving logs, sandboxed execution, error recovery
Deliverables: Open-source artifacts, documented APIs, evaluation reports, performance benchmarks

Latest Updates

Stay up to date with my current work and upcoming events

Current Research

SLM-First Agentic Systems

PhD dissertation research on developing privacy-preserving, cost-efficient agentic AI architectures using Small Language Models. Focus on controller-mediated systems, case-based reasoning, and multi-agent AutoML achieving 10-100× cost reduction with on-premises deployment.

Ongoing Agentic AI

Upcoming Event

Pan African Data Science Workshop

Organizing a comprehensive workshop for the Pan African community covering data ingestion techniques, Hadoop ecosystem (HDFS, Spark, Flink), data wrangling, Power BI, and more.

Coming Soon Workshop

Publication

FICC 2025 Conference Paper

"Analysis of Document Representation for Text Classification" published at the Future of Information and Communication Conference (FICC 2025), Springer. Research comparing LSI and word embeddings for text classification tasks.

March 2025 Springer

Publication

WASSA 2023 Workshop Paper

"Annotating and Training for Population Subjective Views" published at the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis. Collaborative research on belief annotation and classification.

July 2023 ACL Anthology

My Projects

Selected projects in AI, machine learning, and data science

Topic Modeling

Topic Modeling

Exploring hidden topics in large text corpora

Question Answering

Question Answering

Building systems to retrieve precise answers from documents

Image Classification

Image Classification

Potato disease classification using CNN (categorizing images into predefined classes)

Image Segmentation

Image Segmentation

Semantic segmentation—partitioning images into meaningful segments

Predictive Modeling

Predictive Modeling

Building models to forecast trends and outcomes

Data Analytics

Student Success Analytics

Using Power BI and Python to analyze student success metrics and churn

Multi-Modal Learning

Multi-Modal Learning

Exploring text, image, and audio representation for combined learning

Time Series Forecasting

Time Series Forecasting

Financial modeling—predicting future values based on historical patterns

Experience

My professional journey in data science and research

2022 - Present

PhD Candidate - Data Science

Bowling Green State University

Conducting dissertation research on SLM-First Agentic Systems—developing privacy-preserving, cost-efficient AI architectures using Small Language Models. Research focuses on controller-mediated systems, case-based reasoning, and multi-agent AutoML achieving 10-100× cost reduction with on-premises deployment capabilities.

Researching SLM-First Agentic Systems for cost-efficient AI
Developing controller-mediated feature engineering systems
Working on multi-agent AutoML and case-based reasoning architectures
Collaborating with faculty on privacy-preserving AI systems

2022

Data Scientist

ICRISAT

Applied machine learning and computer vision techniques to agricultural data. Developed crop simulation models and implemented deep learning for disease detection.

Built image classification models for crop disease detection
Developed predictive analytics solutions for agricultural outcomes
Created data visualization dashboards for research insights

2021 - 2022

Research Fellow

University of Arizona

Worked in the Computational Language Understanding lab under Dr. Mihai Surdeanu. Focused on natural language understanding and name entity recognition systems.

Developed information extraction algorithms
Implemented question answering systems
Published research on belief extraction

Testimonials

What my colleagues and mentors say about my work

Kim Brooks

"Isaac's work on our student success prediction models has been exceptional. His innovative approach to data analytics has significantly improved our ability to identify at-risk students and provide timely interventions."

Kim Brooks

Director of Student Success Analytics and Technologies, BGSU

Venu G. Dasigi

"As a PhD student, Isaac demonstrates exceptional technical skills and research capabilities. His contributions to our department's research initiatives have been valuable, showing great promise in the field of data science."

Venu G. Dasigi

Professor & Data Science Graduate Coordinator, BGSU

Mihai Surdeanu

"Isaac's work in our NLP lab was characterized by creativity and technical excellence. His contributions to our name entity recognition extraction systems demonstrated both theoretical understanding and practical implementation skills."

Mihai Surdeanu

Head of Computational Language Understanding lab, University of Arizona

Des Johnston

"I had the pleasure of lecturing Isaac on cross-disciplinary topics in quantum mechanics combining mathematical modeling with machine learning. His ability to bridge these fields was impressive and led to novel approaches."

Des Johnston

Faculty Member, Mathematical and Computer Science, Heriot-Watt University

Contact

Reach out for collaborations, research, or a friendly hello

Email

isaackobbyanni@gmail.com