FLUX: Adaptive Intelligence for Everyday Commerce

An AI-powered business intelligence & welfare enablement assistant for India's 10+ million street vendors and micro-entrepreneurs.

🌐 Live Prototype • 📑 API Docs • 🎥 Demo Video • ⚡ Quickstart / Local Run • 🏛️ System Architecture • 📊 Core Capabilities • 📜 API Reference • 🧪 Testing

📌 Submission Overview

Item	Details
Hackathon	OOSC 4.0 Hackathon
Problem Statement	Problem Statement 5: AI for Public Good
Theme	Inclusive AI, Social Impact and Empowerment of Underserved Communities
Sub-Track	AI for Micro-Entrepreneurs and Street Vendors & AI for Accessible Public Services
Repository	github.com/lowkeyd3v/flux
Live Prototype URL	🌐 https://flux-frontend-iojr.onrender.com
Live API Swagger Docs	📑 https://flux-backend-w7fd.onrender.com/docs
Demo Video	Watch 5-Minute Video Demo (YouTube) (Mandatory submission video)
Test Suite	48/48 Automated Tests Passing (Pytest + React Build)

🔗 Live Prototype & API Endpoints

🌐 Live Web Application: https://flux-frontend-iojr.onrender.com
📑 Interactive OpenAPI (Swagger) Docs: https://flux-backend-w7fd.onrender.com/docs
🩺 Live Health & Readiness Probe: https://flux-backend-w7fd.onrender.com/api/health/ready
📊 Live Prometheus Metrics: https://flux-backend-w7fd.onrender.com/api/metrics

🚀 How to Run the Prototype Locally (Alternative)

For evaluators who prefer to run the prototype locally, the entire stack can be launched with a single command using Docker Compose:

# 1. Clone the repository
git clone https://github.com/lowkeyd3v/flux.git
cd flux

# 2. Start the full application stack (PostgreSQL, Backend API, Web Frontend, Prometheus)
docker compose -f docker-compose.prod.yml up -d --build

🌐 Web Dashboard: http://localhost
📑 Interactive OpenAPI (Swagger) Docs: http://localhost:8000/docs
📊 Prometheus Real-Time Metrics: http://localhost:8000/api/metrics
🩺 Health & Readiness Check: http://localhost:8000/api/health/ready

(For step-by-step developer setup without Docker, see Local Developer Setup)

🎥 Demo Video

🔗 Video Submission Link: Watch FLUX 5-Minute Demo Video on YouTube

Video Walkthrough Highlights:

Team Intro & Street Vendor Problem Reality: Context on India's 10M+ street vendors, perishable spoilage, and language barriers.
Live Dashboard & ML Demand Forecasting: Dynamic vendor profiles, daily sales logging, and real-time ML demand inference with confidence bounds.
Weather-Aware Stock Recommendation Engine: Margin-optimized prep recommendations factoring in extreme heat, rain, and perishability.
Grounded Scheme RAG & Trilingual Voice Assistant: Grounded Q&A for PM SVANidhi, MUDRA & Vishwakarma with hands-free voice interaction in Hindi, English, and Hinglish.
Production Engineering, Metrics & 48 Tests: OpenAPI Swagger documentation, live Prometheus telemetry, and 48/48 passing automated tests.
Measurable Social Impact & Closing Remarks: Financial resilience, welfare accessibility, and production readiness.

🌟 Problem Statement & Social Impact

The Reality of Indian Street Vendors

India is home to over 10 million street vendors and micro-entrepreneurs who drive daily urban commerce—chaiwalas, fruit sellers, chaat carts, street tailors, and artisans. Despite their vital contribution, they operate with extreme vulnerability:

Demand Uncertainty & Perishable Spoilage: Vendors rely on guesswork to prep stock. Over-preparation leads to spoiled ingredients and direct financial loss; under-preparation leads to lost daily income.
Weather Sensitivity: Extreme summer heat (>40°C) or sudden monsoon rains decimate footfall, but vendors lack localized forecasting tools tailored to their specific products.
Information Asymmetry in Government Welfare: Schemes like PM SVANidhi (collateral-free credit + 7% interest subsidy), PM MUDRA Yojana, and PM Vishwakarma (₹15,000 toolkits + 5% loans) exist, but complex portals and bureaucratic jargon prevent informal workers from discovering eligibility and applying.
Digital & Language Literacy Barriers: Most vendors prefer speaking in Hindi or colloquial Hinglish over navigating dense English web portals.

💡 The FLUX Solution

FLUX is a lightweight, mobile-first, voice-enabled business intelligence copilot designed specifically for Indian micro-enterprises:

+---------------------------------------------------------------------------------------------------+
|                                       FLUX CORE CAPABILITIES                                      |
|                                                                                                   |
|  [📈 ML Demand Forecast]  --> Predicts unit demand with uncertainty bounds (Random Forest)        |
|  [🌤️ Stock Decision Engine] --> Computes prep quantities capped by budget & adjusted for weather |
|  [🏛️ Scheme RAG Advisor]    --> Answers scheme questions grounded in official government docs     |
|  [🎯 Personalized Matching] --> Matches vendor trades with optimal credit lines & toolkit grants  |
|  [🗣️ Trilingual Voice Assistant] --> Speech-to-Text & Text-to-Speech in English, Hindi & Hinglish|
|  [📊 Production Telemetry]  --> Prometheus metrics, structured JSON logs & distributed tracing    |
+---------------------------------------------------------------------------------------------------+

🏛️ System Architecture

                                  +---------------------------------------------+
                                  |              End User / Vendor              |
                                  |            (Mobile / Desktop PWA)           |
                                  +---------------------------------------------+
                                                         |
                                                         v
                                  +---------------------------------------------+
                                  |     Cloudflare / CloudFront CDN & SSL       |
                                  |      - Hashed Static Chunks Cached 1yr      |
                                  |      - Dynamic API Requests Pass-Through    |
                                  +---------------------------------------------+
                                          /                             \
                       (Static Assets / Web)                    (API Gateway / Reverse Proxy)
                                        /                                 \
                                       v                                   v
+---------------------------------------------+   +-----------------------------------------------------------------+
|          Nginx Alpine Container             |   |                  FastAPI Production Container                   |
|  - SPA Client-Side Fallback Routing         |   |                (Gunicorn + Uvicorn ASGI Workers)                |
|  - Immutable Vite Asset Headers             |   |                                                                 |
|  - Security Headers & Compression           |   |  +-------------------+  +------------------+  +--------------+  |
+---------------------------------------------+   |  | Vendors & Sales   |  | Demand Forecast  |  | Stock Recs   |  |
                                                  |  +-------------------+  +------------------+  +--------------+  |
                                                  |  | Schemes RAG Engine|  | Voice Intent API |  | Health Probes|  |
                                                  |  +-------------------+  +------------------+  +--------------+  |
                                                  |                                 |                               |
                                                  |  [ X-Request-ID Tracing ] [ JSON Structured Logs ] [ Prometheus ]|
                                                  +-----------------------------------------------------------------+
                                                              /                      |                      \
                                                             /                       |                       \
                                                            v                        v                        v
                                            +---------------------+  +----------------------+  +---------------------+
                                            | PostgreSQL Database |  |  External Providers  |  |  Prometheus Scraper |
                                            | (Persistent Volume) |  |  (OpenWeather, LLM)  |  |  & Grafana Dashboard|
                                            +---------------------+  +----------------------+  +---------------------+

🚀 Core Capabilities

1. 📈 Machine Learning Demand Forecasting

Time-Series Random Forest Regressor trained on weekly seasonality, holiday/festival calendar multipliers, price elasticity, and weather sensitivities.
Uncertainty Bounds: Computes 10th percentile (predicted_demand_low), mean point estimate (predicted_demand_point), and 90th percentile (predicted_demand_high) across the decision tree ensemble.
Strict Chronological Validation: Train/validation splits are strictly date-based (80/20 chronological) to prevent future data leakage (Validation MAE ≈ 8.7 units, R² ≈ 0.84).

2. 🌤️ Context-Aware Stock Recommendation Engine

Constraint-Aware Arithmetic: Calculates exact units to prepare based on existing stock (prep_units = forecast - inventory), strictly capped by the vendor's daily working budget (budget / selling_price).
Live Weather Integration: Connects with OpenWeatherMap API for live temperature and rain forecasts (with automatic fallback to neutral baselines when offline).
Explainable Decision Logic: Generates plain-language reasoning and risk scores (low, medium, high) explaining forecast confidence and weather impacts.

3. 🏛️ Government Scheme RAG Assistant (Non-Hallucinating)

Comprehensive Knowledge Base: Curated database of major Indian micro-enterprise schemes:
- PM SVANidhi: Collateral-free working capital loan (₹10,000 → ₹20,000 → ₹50,000) with 7% interest subvention and digital cashback.
- PM MUDRA Yojana: Non-farm micro-loans up to ₹10 Lakhs across Shishu (up to ₹50k), Kishore (₹50k–₹5L), and Tarun (₹5L–₹10L).
- PM Vishwakarma: 18 traditional artisan trades with ₹15,000 toolkit grants and 5% concessional enterprise loans.
- e-Shram Portal: Unorganized worker national database with ₹2 Lakh accidental insurance.
- PMSYM: Old-age social security pension scheme providing ₹3,000/month after age 60.
Vector Retrieval with Lexical Boosting: TF-IDF vector index combined with cosine similarity and domain intent keyword boosting (eligible, documents, subsidy, apply).
Dual-Engine Synthesis: Deterministic, hallucination-free extractive synthesis engine that extracts exact criteria, documents, and portal links, plus optional LLM client integration.

4. 🎯 Automated Personalized Scheme Matcher

Analyzes vendor business attributes (product type, daily budget, location) to automatically surface optimal government initiatives with match reasons and direct application steps.

5. 🗣️ Trilingual UI & Voice Interaction

Instant Trilingual Switching: Seamless toggle between English, हिंदी (Hindi), and Hinglish (conversational Romanized Hindi, e.g., "Aaj kitna banana chahiye?").
Speech-to-Text (STT): Browser-native Web Speech recognition (webkitSpeechRecognition) adapting dynamically between hi-IN and en-IN with zero server latency.
Text-to-Speech (TTS): Integrated audio narration (SpeakerButton) reading recommendations and scheme details aloud in an Indian accent.
Voice Intent Parser: Backend NLP endpoint mapping spoken phrases to actionable platform actions.

6. 📊 Production Observability & Telemetry

Prometheus Metrics (/api/metrics): Exposes request counters, response latency summaries (p50/p90/p99), in-flight requests, database connectivity, and ML/RAG execution timings.
Distributed Tracing: Generates and propagates X-Request-ID across every client request, server log, and database interaction.
Structured JSON Logging: Emits machine-parsable access logs with request timestamps, client IPs, endpoints, status codes, and latencies.
Kubernetes Probes: Standard /api/health/live, /api/health/ready, and /api/health/detailed health endpoints.

🛠️ Tech Stack

Layer	Technologies
Frontend	React 19, Vite, Tailwind CSS, Lucide Icons, React Router v7, Axios
Backend	Python 3.11/3.13, FastAPI, Gunicorn, Uvicorn, Pydantic v2, SQLAlchemy 2.0, Alembic
Database	PostgreSQL 16 (Relational DB with ACID transactions and Alembic migrations)
Machine Learning	Scikit-learn, Pandas, NumPy, Joblib (Random Forest Demand Regressor)
RAG & GenAI	TF-IDF Vector Retrieval, Cosine Similarity, Grounded Extractive Synthesizer, LLM API Client
Observability	Prometheus (`/api/metrics`), Grafana Dashboards, Structured JSON Logging, `X-Request-ID` Tracing
Containerization & CI/CD	Docker Multi-Stage, Docker Compose, Kubernetes (K8s), Google Cloud Run, Nginx Alpine, GitHub Actions
External APIs	OpenWeatherMap API (5-day forecasts & current weather)
Testing	Pytest, FastAPI TestClient, Starlette, AnyIO (48 automated tests)

📁 Repository Structure

flux/
├── .github/
│   └── workflows/
│       ├── ci.yml                    # Automated tests, linting & Docker builds
│       └── deploy.yml                # Automated release & container publishing (GHCR)
│
├── frontend/                         # React Single-Page Application
│   ├── src/
│   │   ├── components/               # UI Components & ErrorBoundary
│   │   ├── pages/                    # Route pages (HomePage, VendorPage)
│   │   ├── services/                 # Axios client with tracing & latency tracking
│   │   ├── utils/performance.js      # Web Vitals & performance telemetry
│   │   ├── translations/             # English, Hindi, Hinglish dictionaries
│   │   └── hooks/                    # Custom React state hooks
│   ├── public/
│   │   ├── _headers                  # Cloudflare Pages / CDN cache headers
│   │   ├── robots.txt                # Search engine crawler configuration
│   │   └── sitemap.xml               # Site URL manifest
│   ├── Dockerfile                    # Multi-stage production container
│   ├── nginx.conf                    # Nginx reverse proxy & cache configuration
│   └── package.json
│
├── backend/                          # FastAPI Backend
│   ├── app/
│   │   ├── main.py                   # App factory with tracing & metrics middlewares
│   │   ├── core/
│   │   │   ├── config.py             # Pydantic Settings & environment config
│   │   │   ├── metrics.py            # Prometheus metrics collector & registry
│   │   │   └── middleware.py         # Request ID tracing & structured JSON logging
│   │   ├── api/                      # REST route controllers
│   │   │   ├── health.py             # Liveness, readiness, detailed & Prometheus endpoints
│   │   │   ├── vendors.py            # Vendor profile CRUD
│   │   │   ├── sales_records.py      # Sales history logging & bulk upload
│   │   │   ├── predictions.py        # Demand forecasting endpoint
│   │   │   ├── recommendations.py    # Stock prep & weather recommendation
│   │   │   ├── schemes.py            # Government scheme RAG & recommendations
│   │   │   └── voice.py              # Voice intent parsing & language catalog
│   │   ├── models/                   # SQLAlchemy ORM models
│   │   ├── schemas/                  # Pydantic request/response schemas
│   │   ├── services/                 # Core business & AI logic
│   │   └── data/schemes_data.json    # Curated scheme knowledge base
│   ├── alembic/                      # Database migration versions
│   ├── tests/                        # 48 Pytest unit, integration & monitoring tests
│   ├── Dockerfile                    # Multi-stage production container
│   ├── docker-entrypoint.sh          # DB migration runner & server starter
│   ├── gunicorn_conf.py              # Production Gunicorn ASGI worker config
│   └── requirements.txt
│
├── deploy/                           # Cloud & Production Deployment Manifests
│   ├── k8s/                          # Kubernetes Manifests (Deployments, Services, HPA, Ingress)
│   ├── cloudrun/                     # Google Cloud Run service definition & deploy script
│   ├── monitoring/                   # Prometheus scrape config & Grafana dashboard JSON
│   ├── render.yaml                   # Render.com infrastructure blueprint
│   └── docker.env.example            # Docker Compose production environment template
│
├── docs/                             # Comprehensive System Documentation
│   ├── PRODUCTION_DEPLOYMENT.md      # Step-by-step production deployment & runbooks
│   ├── CDN_AND_EDGE_GUIDE.md         # CDN edge caching, SSL & performance guide
│   └── MONITORING_AND_OBSERVABILITY.md# Metrics, structured logs, and alert rules
│
├── ml/                               # Machine Learning Pipeline
│   ├── data/                         # Synthetic data generation
│   ├── preprocessing/                # Feature engineering
│   ├── training/                     # Training & evaluation
│   ├── inference/predict.py          # Model loading & inference wrapper
│   └── models/demand_model.joblib    # Serialized Random Forest model
│
├── docker-compose.yml                # Local development database
├── docker-compose.prod.yml           # Full production compose stack
├── LICENSE                           # MIT License
└── README.md

⚡ Quickstart / Local Run Instructions

Prerequisites

Node.js: v18.0 or higher
Python: v3.11 or higher
Docker & Docker Compose (for PostgreSQL database)

Option A: One-Command Production Stack (Docker Compose)

# 1. Clone repository
git clone https://github.com/lowkeyd3v/flux.git
cd flux

# 2. Configure environment
cp deploy/docker.env.example .env

# 3. Start full production stack (PostgreSQL, Backend, Frontend, Prometheus, Grafana)
docker compose -f docker-compose.prod.yml up -d --build

Frontend Application: http://localhost
Backend API: http://localhost/api (or http://localhost:8000)
Prometheus Metrics: http://localhost:9090
Grafana Dashboard: http://localhost:3000

Option B: Local Developer Setup

Step 1: Start PostgreSQL

docker compose up -d

Starts PostgreSQL on localhost:5432 with user flux_user, password flux_password, and db flux_db.

Step 2: Set Up Backend Virtual Environment

cd backend
python -m venv venv

# Activate virtual environment
# On Linux/macOS:
source venv/bin/activate
# On Windows (PowerShell):
.\venv\Scripts\Activate.ps1

# Install dependencies and apply migrations
pip install -r requirements.txt
cp .env.example .env
alembic upgrade head

Step 3: Train ML Demand Model (One-Time)

From the repository root (with virtual environment active):

python -m ml.data.generate_synthetic_data
python -m ml.training.train_demand_model

Step 4: Run Backend Server

cd backend
uvicorn app.main:app --reload --port 8000

API Docs: http://localhost:8000/docs
Health Check: http://localhost:8000/api/health

Step 5: Run Frontend Application

In a separate terminal:

cd frontend
npm install
cp .env.example .env
npm run dev

Web Dashboard: http://localhost:5173

🧪 Running Automated Tests

The test suite covers database persistence, ML model inference, stock preparation logic, weather fallbacks, government scheme vector RAG retrieval, multilingual voice intent parsing, and production observability probes:

cd backend
# With virtual environment activated
pytest -v

Test Suite Summary (48/48 Tests Passing):

tests/test_health.py: Health endpoint and PostgreSQL connectivity checks.
tests/test_monitoring.py: Kubernetes liveness probe, readiness probe, system telemetry, Prometheus metrics, and distributed tracing (X-Request-ID).
tests/test_vendors.py: Vendor CRUD operations and schema validation.
tests/test_sales_records.py: Single and bulk sales logging with constraints.
tests/test_predictions.py: ML prediction inference and uncertainty range bounds.
tests/test_recommendations.py: Stock prep arithmetic, budget caps, risk scoring, and weather fallbacks.
tests/test_schemes.py: Scheme listing, detail lookup, vector retrieval, vendor-context queries, and personalized matching.
tests/test_voice.py: Multilingual intent parsing across English, Hindi, and Hinglish, and supported languages catalog.

📜 API Documentation

All REST endpoints are prefixed with /api (configurable via API_V1_PREFIX). Interactive OpenAPI documentation is available at http://localhost:8000/docs.

Health & Observability Probes

Method	Endpoint	Description
`GET`	`/api/health`	Service health and PostgreSQL database connectivity check
`GET`	`/api/health/live`	Kubernetes liveness probe (checks process responsiveness)
`GET`	`/api/health/ready`	Kubernetes readiness probe (verifies DB, ML model, and RAG data)
`GET`	`/api/health/detailed`	Comprehensive system telemetry, memory stats, CPU, and components
`GET`	`/api/metrics`	Prometheus metrics exposition format (version 0.0.4)

Vendor Profile Management

Method	Endpoint	Description
`POST`	`/api/vendors`	Create a new vendor profile
`GET`	`/api/vendors`	List all registered vendor profiles
`GET`	`/api/vendors/{vendor_id}`	Retrieve specific vendor profile by UUID
`PATCH`	`/api/vendors/{vendor_id}`	Partially update vendor attributes (budget, inventory, price)
`DELETE`	`/api/vendors/{vendor_id}`	Delete vendor profile (cascades to associated sales records)

Sales History Ledger

Method	Endpoint	Description
`POST`	`/api/vendors/{vendor_id}/sales`	Log a daily sales record
`POST`	`/api/vendors/{vendor_id}/sales/bulk`	Batch upload historical sales records
`GET`	`/api/vendors/{vendor_id}/sales`	Retrieve historical sales logs for a vendor
`DELETE`	`/api/vendors/{vendor_id}/sales/{record_id}`	Delete a specific sales record

Demand Forecasting & Smart Recommendations

Method	Endpoint	Description
`POST`	`/api/vendors/{vendor_id}/predict`	ML demand forecast for a target date with weather context
`POST`	`/api/vendors/{vendor_id}/recommend`	Preparation quantity, revenue forecast, risk score & weather summary

Government Schemes & RAG Assistant

Method	Endpoint	Description
`GET`	`/api/schemes`	List all government schemes (with optional `?category=` filter)
`GET`	`/api/schemes/{scheme_id}`	Get full scheme details, eligibility, required docs, and steps
`POST`	`/api/schemes/query`	Natural-language RAG query with source-attributed answers
`GET`	`/api/vendors/{vendor_id}/schemes/recommended`	Personalized scheme recommendations matching vendor profile

Voice & Multilingual Assistant

Method	Endpoint	Description
`POST`	`/api/voice/parse-intent`	Parse spoken vendor query into structured action (schemes, predictions, sales)
`GET`	`/api/voice/supported-languages`	List supported languages (`en`, `hi`, `hinglish`) and speech synthesizer models

🌐 Production Deployment

Ready-to-deploy cloud configurations and manifests:

🐳 Docker Compose Production: docker-compose.prod.yml
☸️ Kubernetes (EKS/GKE/AKS): deploy/k8s/ (backend.yaml, frontend.yaml, postgres.yaml, ingress.yaml, hpa.yaml)
☁️ Google Cloud Run: deploy/cloudrun/ (service.yaml, deploy.sh)
⚡ Render.com Blueprint: deploy/render.yaml
📖 Comprehensive Runbooks:

⚖️ Evaluation Criteria Alignment

Criteria	How FLUX Excels
Innovation	Integrates ML demand forecasting with live weather elasticity, constraint-based inventory arithmetic, trilingual voice assistants, and grounded government scheme RAG into a single unified copilot.
Technical Implementation	Production-ready stack: FastAPI backend, React 19 SPA, PostgreSQL with Alembic migrations, custom TF-IDF RAG engine with lexical boosting, Random Forest with tree variance uncertainty, and Prometheus telemetry.
Feasibility & Social Good	Designed for low digital literacy: trilingual support (Hindi/Hinglish/English), browser-native speech recognition and audio narration, and grounded welfare scheme eligibility for informal street vendors.
Scalability & Architecture	Multi-stage Docker containers, Nginx reverse proxy, Kubernetes Horizontal Pod Autoscalers (HPA), Prometheus metrics scraping, and edge CDN caching.
Code Quality & Testing	Modular architecture, strict separation of concerns, complete typing with Pydantic v2, and 48 automated unit and integration tests running on GitHub Actions CI.

📄 License

This project is licensed under the MIT License.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
backend		backend
deploy		deploy
docs		docs
frontend		frontend
ml		ml
.dockerignore		.dockerignore
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
docker-compose.prod.yml		docker-compose.prod.yml
docker-compose.yml		docker-compose.yml
render.yaml		render.yaml

Latest commit

History

Folders and files

Repository files navigation