CS undergrad (2027) obsessed with one question: how do AI systems actually think at scale?
I build at the intersection of retrieval, reasoning, and real-world impact — from RAG pipelines to multi-step agents. My work leans backend-heavy, systems-first, and always production-aware.
- 🧠 Focus: AI Systems · RAG Architectures · Agent Frameworks · Backend Engineering
- 🔬 Currently building: CodeRAG — an AI debugging agent
- 🌱 Learning: Advanced agentic systems · System design at scale · Hybrid search architectures
- 🤝 Open to: AI/ML collabs, open-source contributions, research discussions
- 📍 Punjab, India
- 🧩 Strong understanding of AI system design (RAG + Agents)
- ⚙️ Ability to build end-to-end backend systems
- 🔍 Focus on problem-solving, not just implementation
- 📊 Experience with real-world datasets and ML workflows
- 🚀 Builder mindset — turning ideas into working systems
AI / ML
Backend
Frontend
Databases & Search
Data Science
Tools & Platforms
An AI-powered debugging system that finds root causes — not just symptoms.
Most debugging tools tell you where the error is. CodeRAG tells you why it happened — by reasoning across your entire codebase context.
User reports bug → CodeRAG indexes codebase + logs + docs + git history
→ Hybrid search retrieves relevant context
→ LangGraph agent reasons across evidence
→ Root cause + suggested fix returned
| Layer | Technology |
|---|---|
| 🔙 Backend | FastAPI |
| 🖥️ Frontend | Next.js |
| 🧬 Code Embeddings | CodeBERT |
| 🗄️ Vector Store | ChromaDB |
| 🔍 Search Engine | Elasticsearch |
| 🤖 Agent Framework | LangGraph |
| 🔀 Search Strategy | Hybrid (BM25 + Vector) |
What makes it different:
- 📁 Code-aware search — understands functions, classes, call graphs
- 🔗 Cross-context reasoning — correlates logs, docs, and commits
- 🧩 Multi-step agents — doesn't guess; it traces
- 💡 Actionable output — fix suggestions with reasoning, not just pointers
class Sneha:
status = "Shipping AI systems that retrieve, reason, and solve real problems"
learning = ["Advanced RAG patterns", "Agentic system design", "Distributed systems"]
exploring = ["Multi-agent orchestration", "Graph-based retrieval", "Eval frameworks for LLMs"]
open_to = "Interesting problems worth solving"- 🔗 Winter of Blockchain — Open-source contributor in Web3 ecosystem
- 🤖 AI/ML Coursework — Deep Learning · NLP · Computer Vision · Generative AI
- 🏗️ Projects shipped — AI debugging systems, web apps, RAG pipelines
- 📈 GitHub Achievements — Pull Shark × 2 · Quickdraw
"Good software is a system, not a script."
I approach every project by asking:
- What breaks at scale? → Design for it upfront
- Where does reasoning fail? → Add structure and evaluation
- What's the actual problem? → Solve that, not the surface symptom
I prefer boring infrastructure that works over clever code that doesn't.
Let's build something that actually matters.