I enjoy building backend systems and understanding how different parts of an application work together.
My background in Mathematics shaped the way I approach engineering:
understand the problem β break it down β build carefully β measure β improve
I started with backend APIs, databases, authentication, testing, caching, and deployment.
Now I am applying that foundation to AI systems, especially Retrieval-Augmented Generation (RAG), where I am building the pipeline from scratch and improving it one version at a time.
An autonomous software debugging agent that diagnoses failing test suites, inspects repository codebases, and applies surgical code edits to repair bugs.
Execution pauses automatically on destructive edits, allowing developers to review syntax-highlighted diffs and authorize changes through an interactive terminal interface before code is written to disk.
Replaces full-file overwrites with character-exact snippet replacement, immediately re-running tests inside an isolated container sandbox to verify each fix without regressions.
Persists an immutable audit trail directly to PostgreSQL, logging the complete unified diff, file path, timestamp, and technical rationale for every applied modification.
Repository: https://github.com/imLeo007/codepilot
A fast & lightweight retrieval engine prioritizing speed and memory.
Document processing runs in the background, allowing uploads to return immediately while parsing, chunking, and storage continue independently.
This takes up to 300mb active ram footprint, Performance was evaluated using Deepevals which resulted in 90% accuracy each time.
The main purpose for this project is that, it can be used as a tool for any AI agent.
Repository https://github.com/imLeo007/docpilot
A heavy weight RAG engine designed to use as a primary source for my retrieval services.
Because of heavy memory footprint and dependency overhead this is flagged as a prototype, it uses Docling which is a production grade framework for document processing.
Since it demands heavy torch dependencies i stripped out all of it and replaced it with small lightweight alternatives that's where i got docpilot.
Repository https://github.com/imLeo007/groundforge
A backend API built to understand how caching can make repeated reads faster while keeping data correct.
The main lesson from this project was simple:
It helped me understand cache hits, cache misses, expiration, invalidation, database consistency, and multi-service backend applications.
Repository https://github.com/imLeo007/profile-cache
Live API https://user-profile-cache-api.onrender.com/docs
A small AI tutoring application where users can sign in, ask questions, and keep their previous conversations.
This is my first rookie project using AI.
The model generates the answer, while the backend handles:
- users
- authentication
- conversation history
- persistence
- application flow
This became the bridge between my backend development work and the RAG systems I started building afterward.
Repository https://github.com/imLeo007/pocket-Tutor
I prefer learning systems from the inside out.
Understand the purpose
β
Design the architecture
β
Reason through the algorithm
β
Build it
β
Debug each layer
β
Measure what happens
β
Improve the real problem
This is why I am building RAG systems from scratch before depending heavily on higher-level frameworks.
My current focus is on building autonomous Agentic workflows.
I am working toward:
Agentic Orchestration
β
Tool Calling
β
RAG
β
Memory
β
HITL
β
Reliable AI Workflows
The goal is not simply to connect models to APIs.
I want to understand how to build AI systems that are useful, grounded, measurable, and reliable.
I am growing toward Backend / AI Engineering, with a long-term interest in building intelligent products that solve real problems.
My main focus is becoming stronger at:
- backend system design
- retrieval systems
- debugging and performance
- AI application architecture
- building AI Agents