From Classroom to Cutting-Edge: Mirza Ilhami Wins Big with LLaMA 3 at Topcoder
At Topcoder, we often say innovation knows no borders—and Mirza Ilhami [mirzailhami] is living proof.
An informatics lecturer by day and a passionate AI innovator by night, Mirza brings a rare blend of academic rigor and real-world application to every challenge he tackles. His journey into AI/ML didn’t start years ago in a lab, but rather earlier this year through a Topcoder challenge—yet his impact has already been significant.
Mirza recently claimed 1st place in the LLaMA Agentic AI Recruitment App and Agents challenge, where he built a powerful, sentiment-aware recruitment tool using models like LLaMA 3 and Claude Sonnet via AWS Bedrock. His solution topped 58 submissions and stood out for its thoughtful use of prompt engineering and modular AI pipelines.
From his motivations and challenge strategies to his dream of building an AI-powered trading platform, this is a story of rapid growth, deep curiosity, and what happens when you merge academic foundations with hands-on innovation.
Let’s dive into the full conversation:
Can you please provide a brief bio of you?
I’m Mirza Ilhami, an informatics lecturer since 2012, teaching programming, web/mobile development, and agile methodologies. I earned my master’s from Binus University, Jakarta, in 2014. I joined Topcoder in 2019, became a Development Copilot in 2021, and started AI/ML work in 2025. My first AI win was the Llama Agentic AI Recruitment App challenge on Topcoder, topping 58 submissions. I’m passionate about coding, AI innovation, and mentoring aspiring developers.
How long have you been working with AI/ML, and how did you get started?
I’ve been working with AI/ML since early 2025. I kicked off with Topcoder’s Llama Agentic AI Recruitment App challenge, which I won. As a software engineer & informatics lecturer, I was eager to explore AI’s potential. Using AWS Bedrock models like Llama3, Claude Sonnet and other models, I built a recruitment solution & other AI challenge, sparking my enthusiasm for AI-driven applications.
What motivates you to take part in challenges like this?
I love tackling real-world problems with cutting-edge tech. The Llama Agentic AI Recruitment App challenge, with 131 registrants, pushed me to innovate as a submitter. Topcoder’s competitive vibe, combined with learning tools like AWS Bedrock, drives me to sharpen my skills and deliver impactful solutions.
How was your experience during the challenge?
The Llama Agentic AI Recruitment App challenge was thrilling but intense, with strict deadlines and 58 submissions from all competitors. I used agile sprints, iterative testing, and robust preprocessing to ensure accuracy. Leveraging Topcoder community feedback helped polish my solution under tight timelines.
What part of the challenge did you enjoy the most?
I loved building the AI agents that powered the recruitment app. Designing prompts for Llama3 to handle candidate interactions was like creating a smart conversational partner. The moment my solution delivered precise, actionable insights was incredibly satisfying, knowing it could streamline hiring in the real world.
What made this challenge particularly difficult for you?
The strict deadlines were the biggest hurdle, with little time to refine the solution. Integrating models (Llama3-70B, Claude-3-7-Sonnet) to deliver consistent results was complex, demanding precise prompt tuning. Ensuring the system was both accurate and fast under pressure stretched my time management and debugging skills.
Can you give a more detailed explanation of your solution? (including code snippets or diagrams)
My winning solution for the Llama Agentic AI Recruitment App used AWS Bedrock to automate candidate screening. I leveraged Llama3-70B for sentiment analysis of interview transcripts, assessing confidence, tone, engagement, and enthusiasm. The system processed noisy data through custom preprocessing and delivered tailored outputs via structured prompts. A modular pipeline ensured scalability.
Here’s the key prompt function for sentiment analysis sent to Bedrock:
def analyze_sentiment(interview_transcript):
prompt = f"""
You are an AI agent for sentiment analysis. Given this interview transcript:
{interview_transcript}
Analyze the sentiment, emotional tone, engagement and enthusiasm in exactly eight or ten concise lines with short summary or explanation of each:
Confidence level: High, Medium, or LowEmotional tone: Positive, Neutral, or NegativeEngagement level: High, Medium, or LowEnthusiasm: High, Medium, or Low
Return the analysis in markdown format within triple backticks (```) with no additional text or explanations.
Start with ``` and end with ```.
"""
return extract_markdown_block(call_llama3(prompt))
A workflow diagram: Transcript Input → Preprocessing → Bedrock Prompt (Llama3-70B) → Sentiment Analysis → Formatted Output. Iterative prompt tuning was critical for precision.
If you could build an AI tool to solve any real-world problem, what would it be?
I’d create an AI-powered trading platform for automated financial investing, accessible to everyone. Using LLMs like Claude-3-7-Sonnet, the AI agent would generate trading strategies, execute trades, and manage risks. It would leverage algorithms like reinforcement learning (e.g., Deep Q-Learning) for backtesting, analyze market trends and sentiment via real-time data, and optimize portfolios for long-term, predictable profits. This could democratize wealth-building, reduce risk, and enhance returns for retail and institutional investors.
Any last thoughts you’d like to add?
I’m thrilled to see AI’s potential in solving real-world problems, from automating recruitment in my winning Llama Agentic AI Recruitment App challenge, or our past challenges like: code reviews (Intelligent Code Review AI Agent System) and converting presentations to videos (PPT to Video AI Converter Web App) and my ongoing AI Agent for Innovation challenge, inspire me to push boundaries. To aspiring AI enthusiasts: jump into these opportunities, experiment boldly, and use platforms like Topcoder to create impactful solutions that transform industries and lives.
--------------------------------------------------------------------------------
Mirza’s journey is a testament to what’s possible when curiosity meets action. In just a few months, he’s gone from experimenting with AI to building award-winning solutions that address real-world challenges. Whether it’s through teaching, competing, or mentoring others, Mirza continues to inspire the next wave of innovators in our community. We’re excited to see what he builds next—and how many others he’ll motivate along the way.
Stay tuned for more stories from Topcoder members turning ambition into innovation, one challenge at a time.