RepLab

Build confidence before adding weight.

RepLab is a mobile-first workout review app that turns a short set video into visual evidence, rep-by-rep feedback, and a smarter next-set recommendation.

Traditional workout logs store what you did. RepLab helps you understand how it looked.

---

Why RepLab?

Most workout trackers are excellent at logging numbers: sets, reps, load, and RIR. But numbers alone do not tell the whole story.

A beginner or intermediate lifter might complete the target reps, but their last reps could become shorter, less controlled, or less consistent. RepLab focuses on that exact moment:

After finishing a set, should I add weight, repeat the same set, slow down, or fix technique first?

RepLab uses a short side-view video to make that decision clearer.

---

Demo preview

Add final screenshots or GIFs here before submitting.

Start	Upload	Review

---

What it does

RepLab guides the user through a focused workout review loop:

1. Choose one supported exercise
2. Learn how to record the set correctly
3. Log basic set context
4. Upload a short side-view video
5. Wait while the set is analyzed
6. Review the rep-by-rep results
7. Get a next-set recommendation

The MVP supports:

• Push-Up
• Bodyweight Squat
• Bicep Curl

Each clip follows a simple recording setup:

• phone placed low near the floor
• phone supported by a water bottle, gym bottle, shoe, or bag
• side-view angle
• full body visible
• decent lighting
• around 3–8 reps
• around 5–15 seconds

---

Core features

Mobile-first RepLab flow

The frontend is designed as a focused mobile app experience:

• Start / Exercise Selection
• Recording Guide
• Set Logger + Video Upload
• Processing / Analyzing
• Rep Lab Review

Visual rep review

The review screen is built around evidence, not chat:

• uploaded set video slot
• rep timeline
• selected keyframes
• best rep vs needs-work comparison
• findings card
• correction cue
• next action card

Computer vision pipeline

The backend processes the uploaded workout video and extracts structured visual information.

The current pipeline:

• receives a video upload through FastAPI
• saves the video temporarily
• runs pose/movement analysis
• tracks exercise-specific joint angles
• detects movement phases
• extracts keyframes for reps
• labels range-of-motion quality
• computes a form score

Gemini multimodal analysis

Gemini is used to analyze extracted rep images and return structured feedback, such as:

• rep scores
• best rep
• worst rep
• correction cards

The backend includes fallback behavior so the demo can remain stable even if external AI configuration is unavailable.

Smart progression recommendation

RepLab turns the analysis into a simple training decision:

• repeat before adding weight
• add reps
• keep the same load
• slow down
• progress when the set is consistent

---

Tech stack

Frontend

• React
• TypeScript
• Vite
• Tailwind CSS
• lucide-react
• Mobile-first UI
• Dark Technical Cyan visual system

Backend

• Python
• FastAPI
• OpenCV
• Pose estimation / movement analysis
• Google Gemini
• Local workout logging

---

Repository structure

HackDays2026/
├── Backend/
│   ├── api/
│   │   └── routes.py
│   ├── cv/
│   │   └── pipeline.py
│   ├── database/
│   │   └── manager.py
│   ├── services/
│   │   ├── gemini_client.py
│   │   └── progression.py
│   └── main.py
│
├── Frontend/
│   ├── src/
│   │   ├── App.tsx
│   │   ├── index.css
│   │   └── services/
│   │       └── api.ts
│   ├── package.json
│   └── vite.config.ts
│
└── docs/
    └── stitch/

---

Running locally

1. Clone the repository

git clone https://github.com/MaCaDo1136/HackDays2026.git
cd HackDays2026

---

Backend setup

From the repository root:

cd Backend
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python -m pip install python-dotenv google-genai pillow uvicorn
python -m uvicorn main:app --reload --port 8000

Check that the backend is running:

http://localhost:8000/health

Expected response:

{
  "status": "ok"
}

---

Frontend setup

Open a second terminal:

cd Frontend
npm.cmd install
npm.cmd run dev

Vite should start at:

http://localhost:5173/

On Windows PowerShell, use npm.cmd instead of npm if script execution is blocked.

---

API overview

Health check

GET /health

Analyze a set

POST /api/analyze_set

Expected FormData fields:

Field	Type	Example
video	File	squat_set_02.mp4
exercise	string	squat
target_reps	number	8
completed_reps	number	8
rir	number	2

Supported exercise values:

UI label	API value
Push-Up	push_up
Bodyweight Squat	squat
Bicep Curl	bicep_curl

Example response shape:

{
  "log_id": 1,
  "form_score": 82.5,
  "progression_recommendation": "Repeat before adding weight",
  "keyframe_paths": ["temp/rep1_good_bottom.jpg"],
  "gemini_analysis": {
    "best_rep": 3,
    "worst_rep": 8,
    "correction_cards": []
  }
}

Logs

GET /api/logs

Returns saved workout logs ordered newest first.

---

MVP scope

RepLab is intentionally focused.

It is not:

• a full workout tracker
• a routine builder
• a social fitness app
• a chatbot wrapper
• a medical tool
• an injury prediction system
• a professional biomechanics platform

It is a focused set-review loop:

Log Set → Upload Video → Review Reps → Get Next Action

---

Design direction

RepLab uses a dark, premium, sports-science lab aesthetic:

• near-black background
• charcoal cards
• Technical Cyan as the primary accent
• Soft Green for good states
• Amber for needs-work states
• Muted Red only for destructive or critical states
• calm, beginner-friendly copy
• replay-style visual analysis

The goal is to make technique review feel practical, accessible, and less intimidating.

---

Challenges

The hardest part was keeping the MVP narrow.

It would have been easy to build a generic AI fitness app, a complete workout logger, or a chatbot that talks about exercise videos. Instead, RepLab focuses on one specific high-value moment:

What should I do after this set?

Other challenges included:

• turning short videos into structured information
• keeping recording requirements simple enough for beginners
• avoiding medical or injury-related overclaims
• making AI support the workflow without becoming the whole product
• balancing visual polish with hackathon time constraints

---

Future improvements

Next steps for RepLab:

• improve rep segmentation accuracy
• support more exercises
• make pose analysis more robust across lighting and camera angles
• compare technique quality across sessions
• personalize progression recommendations
• add editable set history
• show long-term consistency trends
• improve keyframe visualization
• support optional coach explanations without making chat the main product

---

Team

Built during HackDays 2026.

Add team members, roles, and links here.

---

Devpost summary

RepLab turns a short workout set video into a rep-by-rep technique review and a smarter next-set decision, helping lifters build confidence before adding weight.