DMP '25 Week 10 update by Harshit Verma (#380)

Author: therealharshit

public/assets/Images/Pippy_prompt-workflow.png

@@ -0,0 +1,87 @@
+---
+title: "DMP ’25 Week 10 Update by Harshit Verma"
+excerpt: ""
+category: "DEVELOPER NEWS"
+date: "2025-08-16"
+slug: "2025-08-16-dmp-25-therealharshit-week10"
+author: "@/constants/MarkdownFiles/authors/harshit-verma.md"
+tags: "dmp25,sugarlabs,week10,therealharshit"
+image: "assets/Images/c4gt_DMP.png"
+---
+
+<!-- markdownlint-disable -->
+
+# Week 10 Progress Report by Harshit Verma
+
+**Project:** [LLM-powered Debugger for Pippy](https://github.com/sugarlabs/Pippy/issues/95)  
+**Mentors:** [Walter Bender](https://github.com/walterbender), [Ibiam Chihurumnaya](https://github.com/chimosky), [Kshitij Shah](https://github.com/kshitijdshah99)  
+**Assisting Mentors:** [Devin Ulibarri](https://github.com/pikurasa)  
+**Reporting Period:** 2025-08-04 - 2025-08-10   
+
+---
+
+## Goals for This Week
+
+- **Goal 1:** Refine debugger responses to be more accurate and **age-appropriate**.  
+- **Goal 2:** Experiment with a **larger LLM (Gemma-3-27B-it)** for better explanations.  
+- **Goal 3:** Test debugger workflow with various **buggy Python code examples**.  
+
+---
+
+## This Week’s Achievements
+
+1. **Integrated Larger Model (Gemma-3-27B-it)**  
+   - Switched from the previous smaller model to a larger one to generate more coherent and helpful debugging outputs.  
+   - Early results show clearer explanations and smoother response quality.  
+
+2. **Improved Prompt Workflow**  
+   - Refined prompt design to balance **technical accuracy** with **child-friendly language**.  
+   - Ensures that responses are simple, engaging, and encourage a “debugging journey” rather than just spoon-feeding answers. 
+  <img src="assets/Images/Pippy_prompt-workflow.png" alt="Pippy Debugger: Prompt Workflow" width="300">
+
+3. **Extensive Testing with Buggy Code**  
+   - Ran multiple test cases on buggy Python programs.  
+   - Verified that the responses are not only technically correct but also accessible to younger learners.  
+  - Document: [Responses from the LLM-Powered Debugger](https://docs.google.com/document/d/1u5th52avkRwtuu78ojiD2z8llYLuxqbEGOYcBe2-N_8/edit?usp=sharing)  
+
+---
+
+## Challenges & How I Overcame Them
+
+- **Challenge:** Larger models increase inference time.  
+  **Solution:** Optimized API calls and reduced unnecessary context in prompts.  
+
+- **Challenge:** Striking the right balance between correctness and simplicity.  
+  **Solution:** Iterative prompt tuning and trial runs with varied buggy examples.  
+
+---
+
+## Key Learnings
+
+- Learned how to **fine-tune prompt workflows** for both technical correctness and educational clarity.  
+- Understood the trade-offs between **smaller vs larger models** in terms of cost, speed, and output quality.  
+- Gained hands-on experience in **debugging pedagogy** making explanations more interactive for kids.  
+
+---
+
+## Next Week’s Roadmap
+
+- Further refine **debug responses** for edge cases based on mentors feedback.  
+- Integrate full pipeline with Sugar-AI.
+
+---
+
+## Resources & References
+
+**Repository**
+- [Pippy](https://github.com/therealharshit/Pippy/tree/DMP2025/Pippy-Debugger)
+- [sugar-ai](https://github.com/sugarlabs/sugar-ai)
+- [pippy-debugger-server](https://github.com/therealharshit/pippy-debugger-server)
+
+---
+
+## Acknowledgments
+
+Thank you to my mentors, the Sugar Labs community, and fellow contributors for your continued guidance and support!
+
+---