Skip to main content
SYS_INIT // DWARKESH.LAB
DWARKESHRAMANI
Computer Engineering · Full-Stack Engineer
INITIALIZING DWARKESH // BUILD SYSTEM...12%
INDEX / RETURN TO SELECTED WORK
01/Full-Stack Engineer & System Architect·2026·SHIPPED

MiniCode

Competitive programming, rebuilt for real engineers.

Next.js 15FastAPIPostgreSQLRedisGoogle GeminiDockerGitHub REST APITailwind CSS
VIEW SOURCE ON GITHUBLAUNCH LIVE APPLICATION

01 // OVERVIEW

Competitive programming, rebuilt for real engineers.

THE PROBLEM

Traditional competitive programming platforms isolate developers in sterile in-browser textboxes with artificial test runners. They fail to teach git hygiene, modular code structuring, or realistic software engineering workflows, while providing opaque scorecards that offer zero constructive feedback.

THE IDEA & APPROACH

Rebuild the coding arena around authentic Git commits. Students fork challenge repositories, write modular code locally or in Web IDEs, push commits, and receive instant, constructive code critiques powered by Google Gemini and isolated container execution.

02 // SYSTEM ARCHITECTURE & DATA FLOW

Hybrid event-driven pipeline bridging GitHub webhooks to a FastAPI orchestration backend, leveraging Redis queues for asynchronous evaluation and Gemini for semantic code reviews.

FIG 1.0 // PIPELINE ARCHITECTURE (EVENT-DRIVEN WEBHOOK HARNESS)GITHUBGit Push HookNEXT.JS 15App Router HUDFASTAPIAsync OrchestratorREDIS / CELERYTask QueuePOSTGRESQLUser & State DBDOCKERSandbox+ Gemini Judge
DETAILED EXECUTION SEQUENCE
01 →Student pushes commit to private challenge repository
02 →GitHub Webhook triggers FastAPI ingestion endpoint with commit SHA
03 →FastAPI dispatches evaluation task to Celery / Redis queue
04 →Docker container spins up in a hardened sandbox to execute unit test suite
05 →Gemini 2.5 Flash reviews AST, algorithmic complexity, and idiomatic style
06 →Consolidated test telemetry and AI mentorship feedback push live to Next.js frontend

03 // TECHNICAL DECISIONS & TRADE-OFFS

DECISIONDual-mode GitHub webhook ingestion with fallback sandbox
ALTERNATIVES CONSIDEREDGit Integration vs In-Browser Editor
WHY

Fosters industry-standard Git workflow habits (branches, commit messages, PRs) rather than treating code as ephemeral text snippets.

DECISIONFastAPI with AsyncIO over Node.js
ALTERNATIVES CONSIDEREDEvaluation Backend
WHY

Python's robust ecosystem for code parsing, AST manipulation, and seamless SDK interop with Gemini and container runtimes.

DECISIONStructured JSON schema output with Gemini
ALTERNATIVES CONSIDEREDMentorship Feedback Loop
WHY

Ensures evaluation consistency: time complexity, space complexity, security pitfalls, and clean code score without hallucinations.

04 // CHALLENGES & RESOLUTIONS

CHALLENGE // 01

Securing the test sandbox against malicious user code execution (infinite loops, fork bombs, network exfiltration).

CHALLENGE // 02

Handling webhook spikes during live hackathons and competitive sprint windows without dropping submissions.

CHALLENGE // 03

Achieving sub-second UI updates for live leaderboards without polling bottlenecks.

05 // VERIFIED OUTCOMES

Full platform deployed with real-time XP engine, faculty analytics portal, and dynamic roadmaps.

Successfully battle-tested across college coding club sprints with zero submission losses.

Interactive landing page and feature simulator deployed at minicode-intro.vercel.app.

06 // LESSONS & TAKEAWAYS

Asynchronous background tasks are non-negotiable when integrating external LLM APIs in high-concurrency event platforms.
Gamification is vastly more engaging when tied to transparent skill metrics rather than arbitrary point totals.
STATUS: VERIFIED ON GITHUB
INSPECT REPOTEST DEMO