Agentic Pattern Evaluation Study
Budget: $30 – $250 AUD
I’m launching a focused study that compares agentic design patterns. The core goal is to identify viable emerging patterns while grounding every finding in today’s state-of-the-art approaches.
The work unfolds in three tightly linked stages:
1. Critical Review
You’ll map and analyse current agentic architectures, isolating the most influential state-of-the-art patterns and flagging promising, less-documented variations.
2. Prototype Implementation
Selected patterns get translated into small proof-of-concepts with an agentic library—LangGraph is my preferred choice, though I’m open to comparable Python frameworks if a pattern demands it. Each prototype should be minimal yet functional enough to demonstrate key agent flows.
3. Rigorous Evaluation
We will quantify task completion rate, robustness to errors, and operational efficiency for every prototype, applying a common test harness so results stay comparable. Please build a reproducible methodology (data sets, prompts, error-injection strategy, and metrics scripts) and present the findings in clear, visual form.
Deliverables
• Annotated literature & code survey (PDF or Markdown)
• LangGraph-based prototype repo with setup instructions
• Evaluation toolkit (scripts, data, README)
• Comparative results report with discussion of trade-offs and recommendations
Success means the experiments run end-to-end on my machine and the report highlights which patterns excel, where they falter, and why. I’m eager to collaborate on experiment design and can answer domain questions quickly so you can stay focused on building and testing.
The work unfolds in three tightly linked stages:
1. Critical Review
You’ll map and analyse current agentic architectures, isolating the most influential state-of-the-art patterns and flagging promising, less-documented variations.
2. Prototype Implementation
Selected patterns get translated into small proof-of-concepts with an agentic library—LangGraph is my preferred choice, though I’m open to comparable Python frameworks if a pattern demands it. Each prototype should be minimal yet functional enough to demonstrate key agent flows.
3. Rigorous Evaluation
We will quantify task completion rate, robustness to errors, and operational efficiency for every prototype, applying a common test harness so results stay comparable. Please build a reproducible methodology (data sets, prompts, error-injection strategy, and metrics scripts) and present the findings in clear, visual form.
Deliverables
• Annotated literature & code survey (PDF or Markdown)
• LangGraph-based prototype repo with setup instructions
• Evaluation toolkit (scripts, data, README)
• Comparative results report with discussion of trade-offs and recommendations
Success means the experiments run end-to-end on my machine and the report highlights which patterns excel, where they falter, and why. I’m eager to collaborate on experiment design and can answer domain questions quickly so you can stay focused on building and testing.