AI Data Analyst
I evaluate AI responses and write feedback that makes models more accurate and reliable
AI Data Analyst and AI Generalist specializing in response evaluation, RLHF and SFT data, red teaming, and code review in Python and C++.
How I Review
-
01
Read the Prompt First
I start every review by carefully reading the original prompt or instruction. Understanding what was asked is essential to evaluating whether the response actually delivers what the user needed.
-
02
Check Every Claim
I verify facts, test code, and trace reasoning step-by-step. If a response makes a claim, I confirm it against reliable sources or run the code to see if it actually works as described.
-
03
Score Against the Rubric
I apply the project's scoring rubric consistently, rating dimensions like accuracy, completeness, tone, and instruction-following. This ensures my evaluations are objective and aligned with training goals.
-
04
Write Feedback Someone Can Act On
I write clear, specific feedback that explains what went wrong and how to fix it. My goal is to give the model—and the team training it—concrete guidance that leads to measurable improvement.
What I Work On
-
Core Focus
Response Evaluation
I assess AI-generated responses for accuracy, coherence, and alignment with user intent. Every evaluation is grounded in the prompt requirements and scored against clear rubrics to ensure models deliver reliable outputs.
-
Data Quality
RLHF and SFT Data
I create and curate high-quality feedback data for Reinforcement Learning from Human Feedback and Supervised Fine-Tuning. This work directly improves how models learn from human preferences and instructions.
-
Safety Testing
Red Teaming
I probe AI systems for weaknesses, edge cases, and potential failures. By systematically testing boundaries, I help teams identify and fix vulnerabilities before models reach production.
-
Python & C++
Code Review
I review Python and C++ code generated by AI models, checking for correctness, efficiency, and adherence to best practices. My feedback helps models produce code that actually works and follows real-world standards.
Results That Matter
Sample Evaluations
Fact-Check Evaluation
Challenge
An AI response claimed the sky is blue because it reflects the ocean, which is factually incorrect. Additionally, the response exceeded the two-sentence limit specified in the prompt.
Solution
I flagged the incorrect causal explanation and noted that Rayleigh scattering of sunlight is the actual reason. I also highlighted the length violation and referenced the prompt constraint.
Results
The feedback provided a clear, actionable correction that the model could learn from, improving both factual accuracy and instruction-following in future responses.
Code Review Evaluation
Challenge
A Python function to calculate the average of a list would crash when given an empty list, despite the docstring claiming it handled that case gracefully.
Solution
I identified the missing edge-case handling, explained why the code would raise a ZeroDivisionError, and noted the discrepancy between the docstring promise and actual behavior.
Results
My feedback highlighted both the runtime bug and the documentation mismatch, guiding the model toward more robust and honest code generation.
Reasoning Evaluation
Challenge
In a discount calculation problem, the model incorrectly added 25% and 10% to get 35% total discount, arriving at $26 instead of the correct $27 final price.
Solution
I walked through the correct sequential discount calculation: apply 25% first, then 10% to the reduced price. I showed that the two discounts do not simply add and explained the arithmetic error.
Results
The detailed step-by-step correction helped the model understand compound percentage operations, improving its mathematical reasoning on similar problems.
Certifications
I evaluate AI responses and write feedback that makes models more accurate and more reliable. Every review I deliver is grounded in the prompt, verified against facts, and designed to drive measurable improvement.
Skills & Education
Technical Skills
Python, C++, SQL, JavaScript, pandas, LaTeX, Markdown. I use these tools daily for code review, data analysis, and writing structured feedback.
Core Competencies
Critical reading, written feedback, and fact-checking. I break down complex AI outputs and communicate issues clearly to technical and non-technical audiences.
Education
B.Sc. Computer Science, 2023 to 2027, CGPA 4.12 / 5.0. Strong foundation in algorithms, data structures, and software engineering principles.
Let's Work Together
akandemichael268@gmail.com
linkedin.com/in/akande-michael-235a59434
Ready to Improve Your AI Models?
I bring rigorous evaluation, clear feedback, and a track record of measurable results. Whether you need response ranking, code review, or red teaming, I deliver work that makes models more accurate and more reliable. Let's talk about your next project.