╔═══════════════════════════════════════════════════════════════════════════════╗
║ ║
║ CS student. Founder. Engineer who shipped a live AI platform that's ║
║ handling real traffic, earning Google rankings, and running a security ║
║ stack — all solo. Not someday. Right now. ║
║ ║
║ Every project below has real users. Every system below is deployed. ║
║ Every architecture decision below has a reason. ║
║ ║
╚═══════════════════════════════════════════════════════════════════════════════╝
The problem: AI benchmarks are a lie. Vendors cherry-pick prompts, brand recognition skews votes, and nobody can audit results. There is no honest, community-driven evaluation layer.
The solution: Blind head-to-head model battles. Real community prompts. Zero brand visibility until the vote is locked.
┌─────────────────────────────────────────────────────────────────────────────┐
│ SYSTEM ARCHITECTURE │
│ │
│ [User Prompt] │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ Cloudflare CDN + AI Gateway │ ← traffic layer │
│ └────────────────────────┬────────────────────────────┘ │
│ │ │
│ ┌──────▼──────┐ │
│ │ .NET Web API │ ← stateless, horizontally scalable │
│ └──────┬──────┘ │
│ │ │
│ ┌────────────┴────────────┐ │
│ ▼ ▼ │
│ ┌──────────┐ ┌──────────┐ │
│ │ Model A │ │ Model B │ ← identities hidden │
│ │ (blind) │ │ (blind) │ │
│ └────┬─────┘ └────┬─────┘ │
│ └──────────┬────────────┘ │
│ ▼ │
│ [Community Votes] → [Live Leaderboard] → [Open Data] │
└─────────────────────────────────────────────────────────────────────────────┘
What's shipped and running:
▸ Handled 5,000+ requests during peak traffic spike (bots + real users) — CDN absorbed load before origin was touched
▸ Cloudflare AI Gateway as unified LLM routing layer — rate limiting, response caching, per-request observability across all model providers
▸ Real-time security system — login anomaly detection, IP address capture, risk-level classification, automated email alert to account owner on every suspicious event
▸ SEO — ranks #1 on Google for "DualMind Labs", ~Top 95 for "AI arenas"
▸ Full user documentation written, structured, and published
▸ Supabase (PostgreSQL) — chosen deliberately over Firebase. Leaderboard ranking queries require relational SQL at depth. NoSQL would have been unmanageable.
▸ .NET Web API — stateless; horizontal scaling is a config change, not a rewrite
Azure · .NET Web API · Vanilla JS · Supabase · Cloudflare CDN · Cloudflare AI Gateway · GitHub Actions
Three coordinated Flutter apps. One Firebase backend. Real-time payment sync across every role. A production-grade restaurant system where a payment confirmed by the customer appears on the chef screen and admin dashboard in under a second.
┌────────────────┐ ┌────────────────┐ ┌────────────────┐
│ CUSTOMER APP │ │ CHEF APP │ │ ADMIN APP │
│ │ │ │ │ │
│ Browse menu │ │ Receive order │ │ Manage menu │
│ Place order │ │ Update status │ │ View staff │
│ Pay Razorpay │ │ Payment alert │ │ Revenue live │
│ Track live │ │ Order ready │ │ All orders │
└───────┬────────┘ └───────┬────────┘ └───────┬────────┘
└──────────────────────┼──────────────────────┘
▼
┌──────────────────────┐
│ Firebase Firestore │ ← single source of truth
│ + Cloud Functions │ ← Razorpay webhook handler
└──────────────────────┘
▲
┌──────────┴──────────┐
│ Razorpay │
│ Payment lifecycle │
└─────────────────────┘
What's shipped and running:
▸ Razorpay full payment lifecycle — webhook hits Cloud Function, payment status propagates across all three apps instantly, each role sees what they need to see
▸ Real-time order notifications — customer gets confirmation, chef gets order detail, admin sees revenue tick up — all from one Firestore write
▸ Firebase security rules scoped per role — strict access control baked in, not bolted on
▸ CI/CD pipeline via GitHub Actions across all three apps simultaneously
Flutter (3 apps) · Firebase Firestore · Firebase Auth · Cloud Functions · Razorpay · GitHub Actions
Handles routine inbound support volume using Claude / GPT-4o with function calling. Escalates only what genuinely requires human judgment — with full context pre-attached so the human isn't starting cold.
[Inbound query] → [AI agent: classify intent + urgency]
│
┌───────────────┴───────────────┐
▼ ▼
[Resolved] [Escalate]
Auto-respond + log Human review, context
to CRM pre-drafted
Node.js · Anthropic Claude · OpenAI GPT-4o · Supabase · Function Calling · Webhooks
SCALABILITY ──────────────────────────────────────────────────────────────
Cloudflare CDN absorbs spikes before they touch the origin.
The .NET API is stateless — horizontal scaling is a config change.
AI Gateway caches and rate-limits LLM calls; the model layer
isn't hammered unnecessarily.
OBSERVABILITY ────────────────────────────────────────────────────────────
Logging and alerting are day-one architecture, not post-incident patches.
DualMind: real-time login alerts with IP + risk scoring.
AI Gateway: per-request visibility across all model providers.
MealMind: webhook logs and Cloud Function execution traces.
SECURITY ─────────────────────────────────────────────────────────────────
Anomaly detection, IP capture, risk classification, and owner email
alerts were designed into DualMind at launch. Not added after a breach.
TRADEOFFS ────────────────────────────────────────────────────────────────
Firebase vs Supabase:
MealMind → Firebase. Real-time 3-app sync is exactly Firestore's job.
DualMind → Supabase. Leaderboard queries need real SQL. NoSQL fails here.
AI Gateway vs direct API calls:
Unified rate limiting, caching, and observability across providers.
For a multi-model platform, the abstraction pays for itself.
Flutter · Dart · Node.js · .NET · JavaScript · TypeScript · C# · C
Firebase · Supabase · PostgreSQL · MySQL · Azure · Cloudflare · Linux · Docker
GitHub Actions · Git · Next.js · HTML · CSS · Figma · VS Code · Postman
┌─────────────────────────────────────────────────────────────────┐
│ 01 MOBILE APPS → 2–6 weeks │
│ Flutter · iOS & Android · Razorpay/Stripe │
│ Multi-role auth · Firebase · CI/CD · App Store + Play │
├─────────────────────────────────────────────────────────────────┤
│ 02 BACKEND APIs → 1–3 weeks │
│ .NET Web API or Node.js · Auth · Security alerting │
│ Supabase or Firebase · Scalable · Fully documented │
├─────────────────────────────────────────────────────────────────┤
│ 03 AI PRODUCT DEVELOPMENT → 1–4 weeks │
│ RAG · AI agents · Multi-model systems │
│ OpenAI + Claude + Cloudflare AI Gateway · Cost-optimised │
├─────────────────────────────────────────────────────────────────┤
│ 04 FULL-STACK MVPs → 3–8 weeks │
│ Mobile + backend + AI + security + DevOps + deployed │
│ Azure · Cloudflare CDN · CI/CD · Full source handoff │
└─────────────────────────────────────────────────────────────────┘
Upfront scoping. Weekly deliverables. Full source on completion.
Response time < 24 hours. No ghosting. No scope surprises.
01 Ship early. Real users surface problems that planning never will.
02 Design for scale from day one. Stateless APIs. CDN-first.
SQL where queries get complex. These are cheap decisions early.
03 Observability is architecture. Logs, alerts, and dashboards
ship with the product — never added after an incident.
04 Pick the right tool. Firebase for sync. Supabase for SQL.
The tradeoff matters more than familiarity.
05 Own the outcome. The working, monitored, deployed system is
the deliverable. Not the commit.
06 Communicate clearly. Async-first. Unprompted updates. No surprises.
I build systems that handle real traffic, real payments, and real users.
If your team needs an engineer who ships end-to-end and owns the outcome —