✦ Figma Make Challenge ◆ LA Hacks 2026

AI-Powered
Code Review Intelligence

CodeGuard learns your team's unwritten rules from merged PRs and applies them automatically. Powered by AST-level analysis, semantic search, and LLM reasoning.

0
Guidelines Stored
0
PRs Reviewed Today
0
Knowledge Coverage
0
Team Members

How We Used Figma Make

From rapid ideation at 2am to validated architecture — Figma Make helped us think faster and iterate smarter.

💡
Phase 1

Ideation

Prototyped the two-pipeline concept (Evaluate + Commit) visually before writing any code. Aligned the team at 2am.

Architecture alignment at 2am
🎨
Phase 2

Prototyping

Created wireframes showing how an intern's commit flows through CodeGuard: from AST diff to guideline matching to review.

Pitched the idea to team visually
🔄
Phase 3

Iteration

Killed retroactive history mining early, saving 3+ hours. Realized attributes should only accrue from PRs going forward.

Threw away bad ideas early
🚀
Phase 4

Delivery

Created the team breakdown visualization with interfaces and dependencies. Each person could see their scope at a glance.

Team aligned and shipping

The Two Pipelines

CodeGuard runs two distinct pipelines: one learns from merges, the other reviews every PR.

1
📝
PR Opens
GitHub webhook triggers
2
🌳
AST Parse
tree-sitter extracts functions
3
🔍
Filter
Skip whitespace, auto-gen
4
🧠
Embed
Generate vector embeddings
5
🎯
kNN Match
Find k=5-10 similar functions
9
📢
Post Review
Warnings with dismiss option
8
🤖
LLM Check
Claude checks compliance
7
⚖️
Weight
Score by similarity & recency
6
📚
Guidelines
Retrieve from neighbors
tree-sitter SBERT Embeddings ChromaDB Cosine kNN Claude LLM GitHub Actions Numpy Pydantic
1
PR Merges
Triggers learning pipeline
2
🔄
Diff Read
Before/after function states
3
📡
Signal Extract
Touched layers, risky patterns
4
🤔
Worth Check
Is this commit memorable?
7
💾
Store
Write to ChromaDB collection
6
📝
Generate
Structured guideline + reasoning
5
Classify
Positive or negative pattern
Claude Heuristics Git Diff ChromaDB Pydantic

Architecture Arborescence

Click any module to explore its functions and internals.

🛡️
CodeGuard
🗃️

Project Context

Person A
Memory infrastructure — ChromaDB collections, guideline schema, embedding wrapper.
get_collections()
Guideline
write_guideline()
get_embedding()
load_intent()
📚

Commit Pipeline

Person B
Learning pipeline — turns merged commits into reusable guidelines.
is_worth_storing()
classify()
generate_guideline_text()
run_commit_pipeline()
extract_signals()
🔍

Evaluate Pipeline

Person C
PR review engine — AST diff, semantic search, LLM compliance checks.
extract_changed_functions()
filter_functions()
find_similar_functions()
collect_attributes()
run_compliance_check()
⚙️

Integration

Person D
GitHub Actions, PR commenting, dismissal tracking, feedback UI.
post_review_comment()
record_dismissal()
run_codeguard_action()
format_warning()
🗄️ ChromaDB

positive + negative collections

📋 Guideline Schema

pattern, reasoning, diff, confidence

📡 Signal Extraction

touched layers, risky patterns, intent

🔀 Auto / Heuristic

LLM-first or rules-only modes

🌳 tree-sitter

parse, diff_trees, extract bodies

🎯 kNN + Weighting

cosine similarity, threshold, weight

🐙 GitHub Actions

on: pull_request, on: push (main)

❌ Dismissals

Generalized attribute suppression

An Intern's First Week

Follow the path — click each checkpoint to see how CodeGuard transforms a new contributor's journey.

📄
📋
💡
📝
📚
🔮
🏆
💻
🧑‍💼↔ Drag me
🎮 Step 1/5
⚠️
1
🧑‍💻
First Push
📝
Day 1 — First Commit

The Accidental Anti-Pattern

The intern writes code that works, but puts business logic inside a route handler and uses except: to catch all errors. Tests pass, but team conventions are broken.

🔍
2
🤖
AI Reviews
🔬
Day 1 — CodeGuard Evaluates

AST Diff Catches the Pattern

CodeGuard's Evaluate pipeline extracts AST diff via tree-sitter, embeds the changed functions, finds similar code, and flags issues.

"Broad exception handling hides real failures. Keep business logic in service modules."
🔄
3
👩‍🔧
Fix & Learn
🔧
Day 2 — The Fix & The Lesson

CodeGuard Learns from the Merge

The intern fixes the PR. When it merges, the Commit pipeline compares before/after, classifies the fix as positive, and stores the guideline.

"Move route logic to service layer, use specific exception types."
🧠
4
👨‍🏫
Knowledge
📖
Day 5 — Knowledge Compounds

Memory Helps the Next Intern

Another team member makes a similar mistake. CodeGuard now has a stronger guideline from the first intern's experience. Review is faster and contextual.

🚀
5
🦸
Confident!
🏅
Week 2 — Confidence

Contributing with Understanding

The intern understands not just what to change, but why the system works this way. CodeGuard provided the unwritten rules that usually take months to absorb.

📊 Intern Growth
0%

Who Built What

Four people, four roles, clean interfaces between each layer.

🗃️
Person A

Project Context Lead

Memory infrastructure — ChromaDB, guideline schema, embedding wrapper, and repo intent loader.

Key Exports
get_collections()
Guideline
write_guideline()
get_embedding()
ChromaDB Pydantic SBERT
📚
Person B

Learning Pipeline Lead

Turns merged commits into reusable knowledge. Filters, classifies positive/negative, stores structured guidelines.

Key Exports
is_worth_storing()
classify()
generate_guideline_text()
extract_signals()
Claude Heuristics Git
🔍
Person C

Evaluate Pipeline Lead

Runs on every PR. AST diffs, finds similar functions, retrieves guidelines, runs LLM compliance checks.

Key Exports
extract_changed_functions()
find_similar_functions()
collect_attributes()
run_compliance_check()
tree-sitter Numpy Claude
⚙️
Person D

Integration Lead

Wires everything into GitHub Actions, handles PR commenting, dismissal tracking, and the feedback UI.

Key Exports
post_review_comment()
record_dismissal()
run_codeguard_action()
format_warning()
GitHub API Actions YAML