Dodiya Enterprises – Project Architecture & Tools

📐 Full-Stack System Architecture

The project spans four architectural layers: source input (field data), AI mediation (LLM processing), web application (Next.js frontend + backend), and deployment & analytics (Vercel CDN + telemetry). The diagram below maps every component and its connections.

Source / Input Layer

AI / LLM Processing Layer

Frontend / Web App Layer

Backend / Deployment Layer

Analytics / Audience Layer

▼ LAYER 1 — FIELD SOURCE INPUTS

📱 WhatsApp Group

Russian + Sakha
Messages, photos, videos
.txt export files

🌡️ Sensor CSV Logs

Temperature: 23.69°C
RH: 73.35%
75-point soil profile

📷 Media Files

Images (.jpg)
Videos (.mp4)
Field photographs

📺 NVK Yakutia TV

Sakha-language
broadcast coverage
Regional public media

↓

▼ LAYER 2 — AI / LLM PROCESSING ENGINE

🤖 Google DeepMind Antigravity AI (LLM)

Input: WhatsApp .txt + operator instructions (English)
Operations: Discourse segmentation · Register elevation · Bilingual generation · Cultural contextualization · Sakha token preservation
Output: Structured EN + RU web content in JSON format

→

🔄 Transformation Engine

Russian informal → English academic
Sakha phrases → preserved tokens
CSV data → scientific narrative
Latency: < 6 hrs/post

→

📄 JSON Output Schema

date, title, titleRu,
content[], contentRu[],
images[], videos[]

→ activities.json

↓

▼ LAYER 3 — WEB APPLICATION (FRONTEND + BACKEND)

⚛️ Next.js 15 App

React 18 + TypeScript
App Router (SSR/SSG)
Bilingual toggle via useState()

🌐 Activities Page

dodiyaenterprises.com/activities
EN ↔ RU language switch
Image gallery · video embeds

🎨 Tailwind CSS

Responsive design
Mobile-first layout
Utility-based styling

📊 @vercel/analytics

Edge-level telemetry
Geographic attribution
No-cookie compliance

↓

▼ LAYER 4 — VERSION CONTROL + DEPLOYMENT

🐙 GitHub Repository

Main branch
Every post = git commit
Commit b0e8752

→

▲ Vercel CI/CD

Auto-deploy on push
Serverless functions
Global CDN propagation: <3 min

→

🌍 Global CDN

Edge network
dodiyaenterprises.com
HTTPS · Custom domain

↓

▼ LAYER 5 — AUDIENCE & ANALYTICS

🇮🇳 India

342 views · 44.8%
Tech + STEM audience

🇷🇺 Russia

204 views · 26.7%
SPbU · NEFU · ALROSA

🇺🇸 USA

68 views · 8.9%
Climate science

🇰🇿 Kazakhstan

38 views · 5.0%
Post-Soviet networks

🌐 Other

112 views · 14.7%
Circumpolar research

Total: 764 pageviews · 5 nations · Peak: 212 views/day on 19 July 2026 · 73.3% international readership

1 / 4

🔄 The 7-Step LLM Documentation Pipeline

Field Data Collection — Source Corpus Generation

Programme coordinators and students generate field data during the Permafrost School (July 14–26, 2026). Data types: WhatsApp group conversations (Russian + Sakha), JPEG/MP4 media from 53+ images and 4 videos per day, CSV sensor readings, and oral field notes. WhatsApp history exported as .txt file with timestamps.

RussianSakhaCSV

↓

Upload to AI IDE — Context Loading

The operator (Dodiya Enterprises) uploads WhatsApp .txt export to the Antigravity IDE (Google DeepMind). The LLM receives the full conversational context — timestamps, code-switching patterns, image references, cultural vocabulary — in a multi-turn conversation session. Operator provides instructions in English specifying post structure, audience, and preservation requirements for Sakha-language terms.

LLM Context WindowAntigravity IDE

↓

LLM Discourse Analysis & Segmentation

The LLM performs: (a) temporal segmentation — identifying which messages correspond to which day/activity; (b) content extraction — distinguishing logistical messages from scientifically significant events; (c) code-switch detection — identifying Sakha-language tokens for preservation; (d) media association — mapping image filenames to described events.

Transformer ArchitectureAttention Mechanism

↓

Bilingual Content Generation — EN + RU

LLM generates structured bilingual post content per day:
→ English: Formal academic register · international audience · scientific framing · cultural contextualization of NVK, Burnashev heritage, permafrost methods
→ Russian: Formal but accessible · institutional audience (SPbU, ALROSA, families) · bilingual mirror of EN content
Sakha-language terms preserved as authenticated quotations in both language versions.

English outputRussian outputSakha preserved

↓

Operator Review & JSON Structuring

Human operator (mediator layer) reviews LLM output for accuracy, cultural sensitivity, and factual correctness. Approved content is written to src/data/activities.json following the established schema: { date, title, titleRu, content[], contentRu[], images[], videos[] }. Each post entry becomes a structured data object. Media file names are mapped to the corresponding image/video references.

Quality ControlJSON Schema

↓

Version Control & Automated Deployment

Updated activities.json is committed to the GitHub repository (main branch). Vercel's continuous deployment pipeline detects the commit and automatically triggers a rebuild of the Next.js 15 application. The build compiles the updated bilingual content into static and server-rendered pages. Global CDN propagation completes in under 3 minutes. Total pipeline latency: WhatsApp message → live international web post = under 6 hours.

git push → GitHubVercel Auto-Deploy<3 min CDN

↓

International Publication & Telemetry

Content is live at dodiyaenterprises.com/activities. The @vercel/analytics SDK passively captures pageview events at the edge network level with geographic attribution. Readers in India, Russia, USA, Kazakhstan, and other countries access bilingual posts with EN↔RU toggle. Analytics data feeds back into the research documentation as empirical evidence of social impact. Peak engagement: 212 views on July 19, corresponding to the day's scientific field reports.

764 total views5 nations73.3% international

🔁 Key Transformation Examples (WhatsApp → Web)

Source (WhatsApp, Russian)	Output (Web, English Academic)	LLM Operation
"Сегодня лагерь нас посетили НВК и провели съёмки для своей телепередачи"	"NVK Yakutia, the national indigenous-language broadcaster, visited the camp to film a dedicated regional television feature on the students' cryospheric research."	Register elevation + cultural contextualization
куерчэх ытыга наада (Sakha morning expression in WhatsApp)	Preserved as Sakha quotation in italics with contextual gloss in English text	Indigenous language preservation (not translated)
CSV: Temp=23.69°C, RH=73.35%, 75 pts, 120cm pit	"Systematic soil profiling at 75 measurement points across a 120-cm excavation yielded mean ambient readings of 23.69°C at 30cm depth..."	Numeric data → scientific narrative

2 / 4

★ Used in this project ◆ Strong alternative ● Also suitable

Category	Tool / Model Used ★	Strong Alternatives ◆	Key Distinction
AI / LLM Engine	★ Google DeepMind Antigravity AI (Gemini 2.5 Pro backend)	OpenAI GPT-4o · Claude 3.7 Sonnet · Meta Llama 3.3 · Mistral Large 2	Antigravity provides IDE-integrated agentic workflow — reads files, writes code, deploys, all in one session. GPT-4o and Claude excel at long-context multilingual content generation.
AI IDE / Coding Environment	★ Antigravity IDE (Google DeepMind, 2026)	GitHub Copilot + VS Code · Cursor IDE · Windsurf (Codeium) · JetBrains AI Assistant · Replit AI Agent	Antigravity: full agentic loop (plan → code → run → deploy). Cursor: best for code editing with AI context. GitHub Copilot: most widely adopted. Replit: browser-based, no local setup.
Frontend Framework	★ Next.js 15 (React + App Router)	Nuxt 3 (Vue.js) · SvelteKit · Remix · Astro · Gatsby	Next.js: best-in-class SSR + static generation, native Vercel support, large ecosystem. Astro: excellent for content-heavy static sites. SvelteKit: lighter bundle, faster pages.
CSS / Styling	★ Tailwind CSS	Bootstrap 5 · Material UI (MUI) · Chakra UI · Styled Components · Sass/SCSS	Tailwind: utility-first, zero runtime, fastest prototyping. MUI: richest pre-built component library. Bootstrap: most recognized in academia.
Programming Language	★ TypeScript + JavaScript	Python (Django/FastAPI) · Go · Ruby on Rails · PHP (Laravel)	TypeScript: type-safe, excellent for large Next.js apps. Python: dominant for ML/AI backends. FastAPI: ideal if adding Python ML inference endpoints.
Content / Data Layer	★ JSON flat file (activities.json)	Sanity CMS · Contentful · Notion API · Supabase (PostgreSQL) · Firebase Firestore · MongoDB Atlas	JSON file: zero infrastructure, instant setup, AI-writable directly. Sanity/Contentful: headless CMS with rich editor UI, best for non-developer content teams. Supabase: full relational DB with auth.
Translation / i18n	★ LLM-generated bilingual JSON (manual pipeline)	DeepL API · Google Cloud Translation API · i18next library · next-intl · Crowdin	LLM approach: highest quality, cultural nuance, context-aware. DeepL API: best automated MT for EN↔RU. i18next + next-intl: standard i18n routing for Next.js (recommended for scaling to Sakha).
Deployment / Hosting	★ Vercel	Netlify · AWS Amplify · Cloudflare Pages · Railway · Render · GitHub Pages	Vercel: native Next.js support, fastest deploy, built-in analytics. Cloudflare Pages: fastest global edge. Netlify: strong CI/CD for static sites. AWS: enterprise-grade but complex setup.
Version Control	★ GitHub	GitLab · Bitbucket · Azure DevOps	GitHub: largest community, seamless Vercel integration, Actions CI/CD. GitLab: stronger built-in CI/CD pipelines.
Web Analytics	★ Vercel Analytics SDK (@vercel/analytics)	Google Analytics 4 · Plausible Analytics · Fathom · Umami · Matomo	Vercel Analytics: cookie-free, privacy-first, geographic data, zero config. Google Analytics 4: richest feature set. Plausible: open-source privacy alternative.
Document / Paper Generation	★ Node.js scripts (docx + HTML generation)	Pandoc · LaTeX / Overleaf · Python-PPTX · Quarto · R Markdown	Node docx: programmatic DOCX with full control. LaTeX/Overleaf: gold standard for academic publishing. Quarto: modern multi-format scientific document system (highly recommended for future papers).

3 / 4

🤖 AI Model Comparison for This Project's Tasks

AI Model	Provider	Best For	Multilingual	Code Gen	Long Context	Agentic	Project Fit
★ Gemini 2.5 Pro (via Antigravity)	Google DeepMind	Full-stack agentic	✔✔✔	✔✔✔	✔✔✔	✔✔✔	USED — Ideal
GPT-4o	OpenAI	Writing + code	✔✔✔	✔✔✔	✔✔	✔✔	Excellent alternative; weaker agentic file I/O
Claude 3.7 Sonnet	Anthropic	Long-form academic writing	✔✔✔	✔✔✔	✔✔✔	✔✔	Best for research paper drafting; excellent EN/RU
Llama 3.3 70B	Meta (Open Source)	Local / private deploy	✔✔	✔✔	✔✔	✔	Good for on-premise; weaker in Sakha/minority langs
Mistral Large 2	Mistral AI	European lang + code	✔✔✔	✔✔	✔✔	✔	Strong RU/FR support; good open-weight option
DeepSeek-V3	DeepSeek (China)	Code generation	✔✔	✔✔✔	✔✔	✔	Competitive code quality; weaker cultural nuance
Copilot (GPT-4o)	Microsoft/OpenAI	IDE code completion	✔✔	✔✔✔	✔	✔✔	Best for VS Code workflow; no document generation

📋 Project Summary — What Was Built

🌐 Website (dodiyaenterprises.com)

Framework: Next.js 15 + TypeScript + Tailwind CSS
Pages: Home, About, Services, Activities (new), Contact
Activities Page: Chronological bilingual posts (EN↔RU toggle)
Programmes section: "Saint Petersburg State University (SPBU), Alrosa and Mirny Polytechnic Institute" — Permafrost School 2026
Analytics: Vercel SDK — geographic telemetry, cookie-free
Deployment: GitHub → Vercel → global CDN (<3 min)

📝 Research Paper & Conference

Title: "The Sociolinguistic Space of the Peoples of the Arctic as an Object of Innovation-Related Activity"
Authors: Kharlampyeva (SPbU) · Dodiya (India) · Sherstennikova (MPI NEFU)
Conference: VIII Synergy Conference, SPbU, Section 10
Deadline: 1 December 2026 → synergy@spbu.ru
Evidence base: 764 pageviews, 5 nations, 73.3% international
Files: Arctic_Sociolinguistic_Space_v2_FINAL.html (Downloads + Desktop)

🏆 Core Innovation Contributions of This Project

① LLM Documentation Pipeline

First documented case of LLM-mediated real-time conversion of Russian+Sakha WhatsApp field messages into bilingual international academic web content. Publication latency: <6 hours.

② Functional Trilingual Architecture

Engineered sociolinguistic space where Russian (science), Sakha (culture), and English (global reach) perform distinct non-overlapping functions — confirmed by 5-nation audience analytics.

③ Verifiable Social Impact

764 pageviews · peak 212/day · 73.3% international readership. First empirically validated digital documentation of an Arctic science-educational programme using server telemetry as evidence.

🔭 Recommended Next Steps (Future Development)

▶ Add Sakha (third language) toggle to Activities page — LLM-assisted translation + human review

▶ Implement Sanity CMS or Supabase as content layer — enabling non-developer post creation

▶ Use next-intl library for proper i18n routing (EN/RU/Sakha URL paths)

▶ Generate conference PPT (PPTX) from the research paper using python-pptx

▶ Create July 22 Activities post (53 images, 4 videos pending upload)

▶ Submit paper to synergy@spbu.ru by 1 December 2026

4 / 4