const pratyush = {
role : "Full-Stack Engineer & AI Systems Architect",
obsession : "Building AI systems that shouldn't exist yet",
approach : ["Architecture first", "Async by default", "Type-safe everywhere"],
currently : ["Multi-agent reasoning patterns", "LLM evaluation frameworks",
"Real-time streaming architectures"],
superpower : "Turning ambiguous problems into shipped products",
openTo : ["SWE roles", "AI/LLM system design", "Interesting collaborations"]
};|
Novel Multi-Agent Framework with Emergent Role Specialization
🔬 Key Innovations
Stack: |
AI-Powered Job Application Tracking System
🔬 Key Innovations
Stack: |
|
RAG-Powered SOC 2 Compliance Engine
7 Modes: 🔬 Key Innovations
Stack: |
AI-Powered Medical Tourism Platform (Microservices)
🔬 Key Innovations
Stack: |
| Project | What it does | Stack |
|---|---|---|
| 🐛 BetterFeedback | AI feedback analyzer — categorizes Bugs / Features / Pain Points with Gemini sentiment scoring. 22 tests in ~0.14s | Python Flask Pydantic v2 React SQLite |
| ⚙️ text_to_verilog | Natural language → hardware description language (HDL) via LLM | Python LLM |
| 🔍 research-qa | Intelligent document Q&A with semantic search | Python RAG NLP |
| 🌐 Portfolio | Personal site with scroll animations | React Framer Motion |
| Principle | How I Practice It | |
|---|---|---|
| 🏗️ | Architecture First | Design for scale, fault tolerance, and vendor flexibility before writing line one |
| ⚡ | Async by Default | Background queues (BullMQ/Redis), WebSockets, non-blocking pipelines everywhere |
| ✅ | Testing is a Feature | Automated from day one — never retrofitted as an afterthought |
| 🔒 | Type Safety Everywhere | TypeScript end-to-end + Pydantic v2 on all Python services |
| 🔌 | Vendor Flexibility | Swappable AI providers with automatic failover — zero hard lock-ins |
| 🚢 | Ship and Iterate | Real products deployed to real users, then refined with data |
explorations = {
"🤖 AI": "Multi-agent reasoning patterns & advanced LLM orchestration",
"📡 Systems": "Real-time streaming architectures & event-driven pipelines",
"🔐 Security": "Auth patterns, zero-trust design & compliance automation",
"🧪 Evals": "Evaluation frameworks for LLM output quality at scale",
}