review.ai — AI-Powered Code Review Assistant

An AI-assisted code review tool that analyzes pasted code and returns structured, line-level feedback — severity, category, explanation, and a concrete fix — the way a senior engineer would comment on a pull request.

Designed and Developed by

NIKHIL CHARY SRIRAMOJU

BTech CSE (Final Year)

• GitHub: Nikhil-creat
• LinkedIn: nikhil-chary-sriramoju
• Email: sriramojunikhil66@gmail.com
• Instagram: @nikhil__sriramoju
• Facebook: Profile

Why this project

Manual code review is slow and inconsistent, especially for solo developers and students without a team to review their PRs. review.ai gives instant, structured feedback on correctness, style, performance, and readability issues, so developers can catch problems before they ever reach a real reviewer.

Features

• Line-anchored feedback — every finding is tied to an exact line number, shown as a colored marker in the editor gutter
• Severity classification — Critical / Warning / Suggestion / Good, so the most important issues stand out first
• LLM-powered analysis — uses an LLM to reason about correctness and intent, not just pattern-match syntax
• Graceful degradation — falls back to a rule-based heuristic analyzer if the model call fails, so the tool never breaks
• Overall quality score — a 0–100 score summarizing the review at a glance

Architecture

flowchart LR
    A[User pastes code] --> B[Frontend: React UI]
    B -->|POST prompt| C[LLM API]
    C -->|JSON: score, findings| B
    C -.timeout/error.-> D[Heuristic Fallback Engine]
    D --> B
    B --> E[Rendered findings\nline markers + severity cards]

Loading

Flow:

1. User pastes code into the editor panel
2. On "Review code", the app sends the code to an LLM with a structured-output prompt requesting JSON (score, summary, findings)
3. The response is parsed and validated; each finding is mapped to its line number and rendered as an annotation
4. If the model call fails or returns malformed data, a local heuristic analyzer (regex-based checks for common issues like var, ==, unhandled promises) produces a fallback review so the UI always responds

Tech stack

Layer	Technology
Frontend	React (hooks-based, single component)
Styling	Tailwind CSS
AI	LLM API (Claude/OpenAI-compatible /v1/messages schema)
Fallback logic	Rule-based static analysis (regex heuristics)

Project structure

review-ai/
├── code-review-assistant.jsx   # Main React component (editor + findings UI)
├── README.md                   # This file

Running it

This component is self-contained — drop it into any React app with Tailwind configured:

npm install react
# copy code-review-assistant.jsx into your src/ directory
# import and render <CodeReviewAssistant />

To connect a production LLM backend, replace the reviewWithClaude function's endpoint with your own API route (recommended: proxy the request through a backend so your API key is never exposed client-side).

Possible extensions

• GitHub App / webhook integration to auto-comment on pull requests
• Syntax highlighting and multi-language detection
• Persistent review history per user
• Team dashboard aggregating common issues across a codebase

Resume bullet points

Built an AI-powered code review tool that analyzes source code and generates structured, line-level feedback using an LLM, with a rule-based fallback engine ensuring 100% uptime for the review flow.

Designed a JSON-schema-constrained LLM prompting strategy to reliably extract structured findings (severity, category, fix suggestions) from unstructured code review output.

---

Built as a learning project exploring LLM-assisted developer tooling.