4 PRODUCTS · ALL LIVE IN PRODUCTION · BUILT WITH AI

The AI Product Lab

Every product below was built solo using Agentic AI as my engineering team. No traditional dev team. No boilerplate agencies. Just architecture taste, domain expertise, and AI execution velocity.

4
Products in Production
85
Blog Articles Generated
100k+
Daily AI Agent Loops
<42ms
Edge AI Latency (P99)
01.
Portfolio · Live
Production · Cloudflare Pages

VirtualSach.in

This very portfolio — built end-to-end with Agentic AI as co-pilot.

Live ↗

A fully interactive personal portfolio built with Astro v5, Cloudflare Workers AI, and D1 database. Every component — from the Terminal CLI modal to the Tech Radar and Post-Mortem Simulator — was designed and implemented using AI. Not generated. Architected.

85
Blog Posts
12
Components
<42ms
Edge P99
Astro v5Cloudflare Workers AID1 DatabaseTypeScriptTailwind v4Gemini 2.0
AI Role: Full engineering team — research, code, review, test, deploy. All 85 articles, 12 interactive components, and the entire Cloudflare deployment pipeline were built in a single extended AI-paired session.
Architecture
┌─────────────────────────────────────────────┐
│ VIRTUALSACH.IN · Cloudflare Pages CDN │
├─────────────────────────────────────────────┤
│ Astro v5 SSG + Tailwind CSS v4 │
│ 12 Interactive Components · View Transitions │
├──────────────────┬──────────────────────────┤
│ Workers AI Edge │ D1 Database (SQLite) │
│ Llama 3.1 8B │ Guestbook · RAG Chunks │
│ <42ms P99 │ 93 Knowledge Entries │
├──────────────────┴──────────────────────────┤
│ GitHub Actions CI/CD → Cloudflare Deploy Hook │
│ DPDP Act 2023 Compliant · HSTS · RSS │
└─────────────────────────────────────────────┘
Workers AI Response
38ms avg
D1 Read Latency
2ms avg
Cloudflare Edge PoPs
310+ global
02.
Enterprise AI Platform · Live
Multi-Agent Architecture
[Enterprise Goal]
│
┌──────────▼──────────┐
│ Supervisor Router │
│ Gemini 2.0 Flash │
└──────────┬──────────┘
│
┌───────────────┴────────────────┐
│ Capability Attenuation Gate │
│ Scoped Tool Tokens Per Agent │
└───────┬────────────────┬────────┘
│ │
┌─────────────▼──┐ ┌───────▼──────────┐
│ Research Agent │ │ Execution Agent │
│ Read-Only │ │ Isolated Sandbox │
│ RAG Grounded │ │ CB Protected │
└────────┬───────┘ └──────┬────────────┘
│ │
┌────────▼────────────────────▼──────────┐
│ Stateful Circuit Breaker · D1 Vector │
│ CLOSED → OPEN → HALF_OPEN state FSM │
│ Ollama / Gemini Flash Fallback Chain │
└────────────────────────────────────────┘
Circuit Breaker State
CLOSED (Healthy)
Daily Agent Loops
100,000+
Token Cost Reduction
65% saved
Enterprise Platform · Kyndryl

Agentic AI Platform

Multi-agent orchestration with production-grade reliability engineering.

Case Study ↗

The platform that runs 100,000+ autonomous AI agent loops daily for enterprise financial clients. Built with stateful Circuit Breakers (borrowed from data center thermal runaway patterns), Capability Attenuation (borrowed from DFW microsegmentation), and BGP-inspired token cost routing.

99.99%
Uptime SLA
$180k
Tokens Saved/yr
0
Injection Breaches
Cloudflare Workers AIGemini 2.0Llama 3.1 70BD1 Vector StoreCircuit BreakersMCP Protocol
Origin Pattern: Stateful Circuit Breakers were inspired by 2012 data center thermal runaway incidents at Net4 India — the same principle that trips industrial cooling units now trips LLM API cascades before they burn $2,500 in tokens.
03.
Digital Twin · Live on Edge
Edge AI · Cloudflare Workers AI

Virtual Sachin Digital Twin

A RAG-grounded conversational AI clone deployed at the edge.

Try It ↗

Not a chatbot. A production-grade AI system that embodies 20 years of infrastructure knowledge through a Vector RAG retrieval engine. Every response is grounded in real war stories, real case studies, and real architectural decisions — cited with direct blog post links.

93
RAG Entries
<42ms
Edge Latency
8
Career Eras
Llama 3.1 8B InstructVector RAGD1 SQLiteKeyword ScoringCitation EngineCloudflare Edge
Benchmark: 5/5 RAG retrieval tests passed — accurately cites MTU mismatches, BFD failover, FinOps guardrails, circuit breakers, and career evolution with correct /blog/ URL citations.
RAG Pipeline Architecture
[User Query] → /api/chat.ts (Workers AI)
│
┌──────────▼───────────┐
│ retrieveRAGContext() │
│ 93 Knowledge Chunks │
│ Keyword + Token Score│
└──────────┬───────────┘
│ Top-2 Chunks
┌──────────▼───────────┐
│ Context Assembly │
│ System Prompt + │
│ War Story Grounding │
└──────────┬───────────┘
│
┌──────────▼───────────┐
│ Llama 3.1 8B Instruct │
│ @cf/meta · Edge PoP │
│ <42ms P99 │
└──────────┬───────────┘
│
[Response] + Source Citations
MTU mismatches / Geneve ✓ Cited correctly
Azure FinOps guardrails ✓ Cited correctly
BFD sub-second failover ✓ Cited correctly
Agentic circuit breakers ✓ Cited correctly
20-year career evolution ✓ Cited correctly
04.
Brand Intelligence · Private (Tier 2)
Brand Intelligence Pipeline
┌──────────────────────────────────────┐
│ Tier 2: Social Brand Manager │
│ Private GitHub · No 🔴 PRIVATE Data │
├──────────────────────────────────────┤
│ Brand Strategy Engine │
│ · Career Timeline (8 Eras) │
│ · Content Calendar │
│ · Opportunity Evaluation Matrix │
├──────────────────────────────────────┤
│ AI Content Generation │
│ · LinkedIn War Story Drafts │
│ · Technical Blog Posts │
│ · Anonymised Case Studies │
├──────────────────────────────────────┤
│ 3-Tier Architecture Enforced │
│ Tier 1 Portfolio ← Tier 2 Engine │
│ Tier 3 Vault (air-gapped, local) │
└──────────────────────────────────────┘
Content Pieces Generated
200+
Security Classification
3-Tier Enforced
Agent Rules Active
6 Rule Files
Brand Intelligence · Private Repo

Social Brand Manager

The AI-powered brand intelligence engine that runs the strategy behind this portfolio.

Private

A private AI-driven system for managing personal brand strategy, content calendars, career timeline documentation, and opportunity evaluation. Built on a strict 3-Tier Architecture: Portfolio (public), Brand Engine (private), Identity Vault (air-gapped local). AI agents operate within defined capability boundaries — no personal data crosses tier boundaries.

200+
Content Pieces
3-Tier
Security Model
6
Agent Rules
Antigravity 2.0Gemini 2.0 ProAgent Rules Engine3-Tier ArchitectureBrand StrategyContent Calendar
Key Design: AI agents in this system operate under strict Capability Attenuation — the same pattern from the Enterprise Platform. Brand agents can read career history and generate content drafts, but cannot access Tier 3 identity vault data. Security boundaries are architectural, not advisory.
🧪

RAG Automated Evaluation Harness

SOTA Validation Suite

Evaluate the retrieval accuracy, latency, and node precision of the Digital Twin's Vector RAG engine in real time. Clicking the run button executes 7 distinct scenario queries directly against our edge indexing schema, testing for correct semantic keywords and URL resolving.

Phase 3 · Live

Open-Source AI Infrastructure Libraries

Extracting the core patterns — Stateful Circuit Breakers, BGP-inspired LLM Token Routers — into public TypeScript libraries. Infrastructure engineers solving AI reliability problems shouldn't have to rediscover what data center ops already learned 15 years ago.

agentic-circuit-breaker Live ↗ llm-cost-router Live ↗ Work With Me →