Deshi Digest logoDeshi Digest Docs
Google Gemini API BuildFest 2026 Submission

Deshi Digest public documentation

A culturally-aware nutrition companion for Bangladesh that combines local food knowledge, API-grounded chat, plate analysis, user profiles, and Supabase-backed meal history.

Live route

/docs

Access

Public

Core AI

Hybrid AI Stack

Database

Supabase

Problem

Nutrition advice rarely fits daily Bangladeshi life

Generic diet apps often ignore rice, dal, local fish, vegetables, and everyday Bangladeshi meal patterns.
Many recommendations assume expensive imported foods instead of budget-aware, locally available choices.
People need safer guidance that explains tradeoffs without pretending to replace a doctor or dietitian.

Solution

Nanumoni gives practical, culturally-aware guidance

Product Overview

Deshi Digest is a nutrition companion built around Bangladeshi food culture. It helps users chat about meals, analyze plates, track history, and receive practical suggestions grounded in local ingredients and responsible AI boundaries. The system uses Gemini for friendly language only, Edamam for image food detection, and database/API retrieval for Bangladeshi food knowledge, Supabase-backed food and user data, model routing for AI tasks, and a custom model training direction for improving culturally aware nutrition guidance over time.

Warm conversational diet guidance
Vision-based plate analysis
Profile-aware recommendations
Meal and chat history

Product Demo

Working app areas available from the live build

Chat with Nanumoni

Ask nutrition questions and receive contextual guidance.

Open route

Analyze My Plate

Upload or capture food images for Edamam-assisted detection and nutrition database estimates.

Open route

Dashboard

Review user profile context, meal history, and personalized summaries.

Open route

Tech Stack

Implementation stack

ReactTanStack StartTypeScriptVercel Server FunctionsSupabase AuthSupabase PostgresSupabase StoragepgvectorGoogle Gemini APIAI SDKTailwind CSS

AI Models

Gemini model usage

Gemini 2.5 Flash-Lite

Natural-language conversation and friendly explanations over retrieved data.

Data Sources

Food, profile, and external reference data

Internal and curated data

  • Curated Bangladeshi food knowledge base stored in application code and Supabase.
  • FCTB-based food context and local nutrition references.
  • User profile, chat threads, meal logs, and plate analysis records in Supabase with RLS.

External reference APIs

  • USDA FoodData Central for nutrition lookup support.
  • RxNorm for medication reference context.
  • WHO ICD references for medical classification context where appropriate.

Architecture

System architecture

Frontend React/TanStack
Vercel Server Functions
Gemini explanation / Supabase / External APIs
Supabase Postgres + pgvector
flowchart LR
  A[Frontend React/TanStack] --> B[Vercel Server Functions]
  B --> C[Gemini explanation / Supabase / External APIs]
  C --> D[Supabase Postgres + pgvector]

Data Flow

How a request moves through the product

1

User submits chat, profile, or plate input from the React UI.

2

TanStack Start server functions attach auth context and validate request shape.

3

External APIs and Supabase provide lookup data; Gemini only provides optional friendly wording.

4

Results and user history are returned to the UI and persisted through Supabase policies.

Responsible AI

Safety boundaries are explicit

Health guidance limits

Nanumoni gives nutrition education and practical meal guidance. It does not diagnose, prescribe, or replace qualified medical advice. Higher-risk health questions should be escalated to clinicians.

Mitigations in scope

  • Clear not-medical-advice language in user-facing AI workflows.
  • Profile-aware but conservative recommendations.
  • RLS-backed storage for user-specific data.
  • Cultural sensitivity around Bangladeshi food patterns and affordability.

Admin Editing

Admin controls roadmap

A full WYSIWYG documentation editor is not implemented yet. The route is prepared for simple controlled visibility through a config object, and the next admin milestone is an editable docs dashboard.

Visibility toggle prepared through docsConfig.isPublic
Scheduling config prepared with availableFrom and availableUntil
Future editable docs dashboard planned

Search and Export

Documentation roadmap

Full documentation search and PDF export are planned but not shipped in this submission build. This page keeps the scope honest by presenting the live documentation content directly instead of showing non-working controls.

Planned: full-text docs search
Planned: PDF export for submission packets

Roadmap

Product roadmap

Admin editable documentation dashboard
Full-text documentation search
PDF export for judges, mentors, and partners
Expanded verified Bangladeshi food database
Deeper physician and dietitian review workflows
Localized market and vendor-aware recommendations

Team

Team showcase

Changelog

Recent project changes

Public docs module upgraded for submission

Added access scheduling config, team showcase, architecture, data flow, responsible AI, and roadmap sections.

Backend integration completed

Supabase auth, database migrations, server functions, Gemini integration, and API health checks were added.

Core app flows assembled

Landing page, dashboard, chat, onboarding, plate analyzer, and nutrition guidance screens were prepared.