{"id":51262957,"url":"https://github.com/swayam-mishra/hackerrank-orchestrate","last_synced_at":"2026-06-29T13:32:05.835Z","repository":{"id":367705039,"uuid":"1262148244","full_name":"swayam-mishra/hackerrank-orchestrate","owner":"swayam-mishra","description":"HackerRank Orchestrate hackathon AI agents: a multi-modal damage-claim verifier (3rd of 1,773) and a support-triage RAG agent (60th of 1,349). The model describes; plain code decides","archived":false,"fork":false,"pushed_at":"2026-06-27T08:09:15.000Z","size":10810,"stargazers_count":1,"open_issues_count":0,"forks_count":1,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-27T09:07:14.175Z","etag":null,"topics":["ai-agents","hackathon","llms","multimodal","prompt-engineering","rag"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/swayam-mishra.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","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":"2026-06-07T16:28:14.000Z","updated_at":"2026-06-27T08:09:18.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/swayam-mishra/hackerrank-orchestrate","commit_stats":null,"previous_names":["swayam-mishra/hackerrank-orchestrate"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/swayam-mishra/hackerrank-orchestrate","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/swayam-mishra%2Fhackerrank-orchestrate","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/swayam-mishra%2Fhackerrank-orchestrate/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/swayam-mishra%2Fhackerrank-orchestrate/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/swayam-mishra%2Fhackerrank-orchestrate/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/swayam-mishra","download_url":"https://codeload.github.com/swayam-mishra/hackerrank-orchestrate/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/swayam-mishra%2Fhackerrank-orchestrate/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34929698,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-29T02:00:05.398Z","response_time":58,"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":["ai-agents","hackathon","llms","multimodal","prompt-engineering","rag"],"created_at":"2026-06-29T13:32:05.204Z","updated_at":"2026-06-29T13:32:05.817Z","avatar_url":"https://github.com/swayam-mishra.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# HackerRank Orchestrate: my hackathon projects\n\nThis repo holds my entries to **HackerRank Orchestrate**, a monthly 24-hour event where you design\nand ship an AI agent. Each month lives in its own folder with its own data, code, and a single README\nthat walks through the whole thing. If you're here to learn from the code, start with whichever\nproject sounds more interesting and read its README top to bottom.\n\n| Month | Project | What it does | Model | Result |\n|---|---|---|---|---|\n| **June 2026** | [Multi-Modal Damage-Claim Verification](./hackerrank-orchestrate-june/) | Looks at the photos on a damage claim and decides whether they back up the claim, contradict it, or aren't enough to tell. Fills in 14 explained columns. | Claude Opus 4.8 (vision) | 🥉 **3rd of 1,773** |\n| **May 2026** | [Support Triage Agent](./hackerrank-orchestrate-may/) | Reads support tickets for three help desks (HackerRank, Claude, Visa) and either answers them from local docs or sends them to a human. Fills in 5 columns. | Claude Haiku 4.5 | **60th of 1,349** |\n\nEach project keeps its own scraped final leaderboard:\n[`leaderboard-june.json`](./hackerrank-orchestrate-june/leaderboard-june.json) and\n[`leaderboard-may.json`](./hackerrank-orchestrate-may/leaderboard-may.json).\n\n---\n\n## June 2026: Multi-Modal Damage-Claim Verification\n\nThe whole project rests on one idea: **the vision model only describes what it sees, and plain code\nmakes the actual decision.** Keeping those two jobs apart pays off in a few ways. You get the same\nanswer every time (the decision is just code, so you can save the model's observations and re-run the\nlogic offline with no API calls). It's auditable, since every column traces back to a specific rule.\nAnd it resists trickery: an instruction hidden inside an image can confuse one observation but can\nnever reach the code that decides. A person's claim history can add caution, but the structure makes\nit impossible for history to flip a verdict on its own.\n\n**[Read the June README](./hackerrank-orchestrate-june/README.md)** for the architecture, the\ncommands, the decisions, and the results.\n\n## May 2026: Support Triage Agent\n\nA command-line agent that answers or escalates support tickets using only a local set of help docs,\nwith no internet calls while it runs. Each ticket goes through a pipeline: filter out junk, clean up\nthe search query, search the docs (fast keyword search, then a small local model re-orders the best\nhits), score the confidence, check a few hard escalation rules, have Claude draft a reply, validate\nand if needed repair it, strip anything the model made up, and log the reasoning. As in June, the\nmodel only writes; the routing decisions are plain code you can read.\n\n**[Read the May README](./hackerrank-orchestrate-may/README.md)** for setup, a file-by-file guide,\nthe decisions behind it, and the results.\n\n---\n\n## How this repo is organized\n\n```\n.\n├── hackerrank-orchestrate-june/   June project (code/, dataset/, README.md, leaderboard-june.json)\n├── hackerrank-orchestrate-may/    May project  (code/, data/, support_tickets/, README.md, leaderboard-may.json)\n├── .gitignore                     one shared ignore file for the whole repo\n└── README.md                      you are here\n```\n\n## Good to know\n\n- There's **one `.gitignore` at the top** covering both projects (secrets, virtual environments,\n  caches, and generated run files). Each project keeps its own `requirements.txt` and `.env.example`.\n- **Secrets stay out of git.** The `.env` files are ignored and never committed. To run either agent,\n  copy that project's `.env.example` to `.env` and paste in your own `ANTHROPIC_API_KEY`.\n- The challenge **datasets, problem statements, and rulebooks belong to HackerRank.** Each project's\n  README says what's included and what isn't.\n- **License.** My code is released under the [MIT License](./LICENSE). That covers my code only, not\n  HackerRank's problem statements or any bundled help-center data, which stay their owners' property.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fswayam-mishra%2Fhackerrank-orchestrate","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fswayam-mishra%2Fhackerrank-orchestrate","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fswayam-mishra%2Fhackerrank-orchestrate/lists"}