Ghost: An Open‑Source AI‑Powered Investigation Platform

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

Ghost is a web‑based platform that lets developers and investigators automatically surface, enrich, and narrate code‑base insights. By ingesting a repository’s commit history, file changes, and issue metadata, Ghost generates concise evidence + finding reports and a readable narrative that can be shared with stakeholders or used for audit trails. The UI blends a sleek dark‑mode layout with glass‑morphism cards, animated progress bars, and smooth hover effects, giving users a premium, “data‑visualization meets AI‑assistant” experience.

GitHub url - https://github.com/Praveen-Raj-Thulasi/Ghost

AI Engine – Leveraged the open‑weight Ollama Llama 3.1 8B model (self‑hosted) via a thin wrapper (ollamaProvider.js) that streams token‑level completions for evidence extraction and narrative generation. Backend – Node.js/Express server (server/src/app.js) with a modular service layer: repositoryService parses git logs, file diffs, and repository metadata. analysis services (activityAnalyzer, projectMetadata, unfinishedWorkScanner) turn raw data into structured evidence objects. ai providers (ollamaProvider, evidenceOnlyProvider) supply context‑aware completions; a responseValidator ensures JSON‑compatible outputs. Frontend – Vite‑powered React app (client/src) styled with vanilla CSS enriched by dynamic gradients, Google‑Font Inter, and subtle micro‑animations on card hover and loading spinners.

DevOps – Docker‑compose stacks for local Ollama, Postgres (for persisted evidence), and the Node server; CI pipelines run linting, unit tests, and auto‑deployment to Railway. Open‑Source Stack – All dependencies are MIT/BSD‑licensed; the codebase lives entirely on GitHub, enabling anyone to fork, extend, or host their own instance.

Open innovation lets Ghost tap into the thriving ecosystem of community‑maintained LLM runtimes (Ollama, vLLM, LMStudio) and freely‑licensed data‑processing libraries. Because the platform avoids closed‑API lock‑in:

Transparency – Researchers can audit the exact prompts, model versions, and token‑level outputs, fostering reproducibility. Extensibility – Teams can swap in a newer open‑weight model or a custom fine‑tuned checkpoint without renegotiating contracts or paying per‑token fees. Cost‑Effectiveness – Running models locally eliminates API costs, making large## Title Ghost: An Open‑Source AI‑Powered