{"id":28288407,"url":"https://github.com/shchoholiev/assets-manager-start-projects-evaluation","last_synced_at":"2026-05-09T02:32:26.191Z","repository":{"id":290068766,"uuid":"973276202","full_name":"Shchoholiev/assets-manager-start-projects-evaluation","owner":"Shchoholiev","description":"Evaluating OpenAI models on selection of code snippets for a project description.","archived":false,"fork":false,"pushed_at":"2025-05-24T20:10:09.000Z","size":368,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"master","last_synced_at":"2025-09-19T23:30:48.508Z","etag":null,"topics":["chain-of-thought","few-shot-learning","model-evaluation","openai","prompt-engineering","python","zero-shot-learning"],"latest_commit_sha":null,"homepage":"","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Shchoholiev.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-04-26T16:37:22.000Z","updated_at":"2025-05-24T20:10:13.000Z","dependencies_parsed_at":null,"dependency_job_id":"54a17a73-bcb4-4310-b9d7-3e38b2745f4a","html_url":"https://github.com/Shchoholiev/assets-manager-start-projects-evaluation","commit_stats":null,"previous_names":["shchoholiev/assets-manager-start-projects-evaluation"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/Shchoholiev/assets-manager-start-projects-evaluation","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shchoholiev%2Fassets-manager-start-projects-evaluation","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shchoholiev%2Fassets-manager-start-projects-evaluation/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shchoholiev%2Fassets-manager-start-projects-evaluation/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shchoholiev%2Fassets-manager-start-projects-evaluation/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Shchoholiev","download_url":"https://codeload.github.com/Shchoholiev/assets-manager-start-projects-evaluation/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shchoholiev%2Fassets-manager-start-projects-evaluation/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":32804940,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-08T08:22:46.396Z","status":"online","status_checked_at":"2026-05-09T02:00:06.633Z","response_time":123,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["chain-of-thought","few-shot-learning","model-evaluation","openai","prompt-engineering","python","zero-shot-learning"],"created_at":"2025-05-22T00:11:53.407Z","updated_at":"2026-05-09T02:32:26.175Z","avatar_url":"https://github.com/Shchoholiev.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"# assets-manager-start-projects-evaluation\nThis repository contains the code and data used to evaluate different AI models and prompting techniques for selecting relevant code snippets based on project descriptions. Selection algorithm is implemented in [assets-manager-api](https://github.com/Shchoholiev/assets-manager-api)\n\n## Table of Contents\n\n- [Overview](#overview)  \n- [Model Selection Evaluation](#model-selection-evaluation)  \n- [Prompt Engineering Evaluation](#prompt-engineering-evaluation)  \n- [Results](#results)  \n- [Getting Started](#getting-started)  \n- [License](#license)  \n\n## Overview\n\nAssets manager API automatically assembles starter projects for digital banking components by selecting the most relevant code snippets based on a text description. I compared several AI models by processing 100 synthetic “projects” against a pool of 100 candidate snippets, measuring how accurately each model retrieved exactly the needed fragments.\n\n## Model Selection Evaluation\n\nModels are benchmarked in ascending order of cost (per 1 M in + 1 M out tokens):\n1. **gpt-4.1-nano** ($0.50)  \n2. **gpt-4o-mini** ($0.75)  \n3. **o3-mini** ($5.50)  \n4. **o1** ($75)  \n5. **gpt-4o** ($12.50)  \n6. **gpt-4.5** ($225)\n\nEach model received the project description and the full snippet pool and returned its top choices. I classified outcomes into:\n- **Exact match**: all and only needed snippets selected  \n- **Partial – Extra**: most needed, but with extras  \n- **Partial – Missing**: missed some needed  \n- **Partial – Mixed**: extras and missing together  \n- **Mismatch**: irrelevant selections  \n\n![Model Comparison](images/model_comparison.png)\n\n## Prompt Engineering Evaluation\n\nUsing **gpt-4o-mini** as my chosen backbone, I tested three prompting styles on the same data:\n\n1. **Zero-shot** (no examples)  \n2. **Few-shot** (with example pairs)  \n3. **Chain-of-Thought** (model explains reasoning then selects)\n\n![Prompting Techniques Comparison](images/prompting_comparison.png)\n\n| Technique           | Exact Match |\n|---------------------|------------:|\n| Zero-shot           | 16%         |\n| Few-shot            | 22%         |\n| Chain-of-Thought    | 34%         |\n\nChain-of-Thought prompting achieved near the performance of much costlier models at a fraction of the price.\n\n## Results\n\n- **Best Model**: `gpt-4o-mini` for cost/quality trade-off  \n- **Best Prompting**: Chain-of-Thought  \n\n---\n\n# Notes\n\nTo access API endpoints exposed by a different docker container from within a current one, use `host.docker.internal` instead of `localhost`","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fshchoholiev%2Fassets-manager-start-projects-evaluation","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fshchoholiev%2Fassets-manager-start-projects-evaluation","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fshchoholiev%2Fassets-manager-start-projects-evaluation/lists"}