# Hi, I'm Safal Gupta
**AI Engineer · ML Systems · Applied Research · Software Engineering**
I build AI systems from first principles — from models and retrieval infrastructure to agents, developer tools, and production software.
Currently studying Computer Science at **VIT Vellore** and exploring the engineering behind modern AI systems.
---
### What I'm working on
- Building **LLM systems, RAG pipelines and AI agents**
- Studying **vector databases, ANN search and retrieval systems**
- Exploring **efficient ML systems and Edge AI**
- Building infrastructure-oriented projects in **distributed systems**
- Reproducing and implementing ideas from recent AI research papers
- Contributing more deeply to open-source AI infrastructure
---
### Areas I'm interested in
```text
Machine Learning
Deep Learning
LLMs
AI Agents
Retrieval Systems
Vector Databases
Distributed Systems
Edge AI
ML Infrastructure
AI Systems ResearchAn intelligent workspace manager for developers working on storage-constrained machines.
Automatically understands projects, archives inactive work, manages local storage and keeps development environments lightweight.
AI Agents System Design macOS Developer Tools
A distributed database project built to understand how modern databases work internally.
Exploring:
Client → Router → Shards → Raft → WAL → SSTables
Focused on replication, consensus, durability and distributed storage.
A real-time 3D commerce engine for jewellery.
Built around physically based rendering, realistic metals, gemstone optics and interactive product customization.
TypeScript WebGL Three.js PBR
Built during my internship at JCB India.
Designed production systems for SOP management, approval workflows and industrial process digitization.
Next.js PostgreSQL Prisma Redis Power Platform
Instead of only using AI libraries, I'm trying to understand how the underlying systems work.
Some things I'm studying/building:
Tokenizer
↓
Embeddings
↓
Vector Search
↓
Retrieval
↓
Transformer
↓
LLM
↓
Tool Calling
↓
Agent
↓
Evaluation
↓
Production AI System
My goal is to be able to build, debug and optimize every layer.
Languages
Python C++ Java Rust Go TypeScript
AI / ML
PyTorch scikit-learn Transformers LLMs RAG Vector Search
Systems
PostgreSQL Redis Docker Linux Git AWS
Web
Next.js React Node.js
- Retrieval-Augmented Generation
- Approximate Nearest Neighbor Search
- Agent memory
- Efficient inference
- Small language models
- Multimodal AI
- Edge AI
- Distributed ML infrastructure
- AI evaluation
- Long-context systems
I'm spending this phase of my career doing one thing:
Build enough serious systems that my GitHub explains my engineering ability before my résumé does.
LinkedIn · X / Twitter · Portfolio
A few things would make this substantially stronger.
**Don’t add** dozens of technology icons, skill-progress bars, visitor counters, snake animations, random trophies, or huge GitHub-stat cards. They make student profiles look busy rather than technically strong.
Instead, your pinned repositories should eventually tell this story:
1. **`llm-from-scratch`** — tokenizer → transformer → inference
2. **`vector-engine`** — build ANN/vector search yourself
3. **`mini-spanner`** — distributed database
4. **`ai-workspace`** — your developer storage/memory agent
5. **`paper-reproductions`** — implementations of recent AI papers
6. **one excellent production AI product**
That combination creates a much stronger impression than having twenty ordinary ML projects such as Titanic, house-price prediction, or a basic movie recommender.
Your profile should make someone think:
> “This person doesn't just call AI APIs. He wants to understand how the system underneath them works.”
That is the identity I would build your GitHub around.