Hi, I'm Sujith Putta.
I shape digital systems as an
Generative AI Developer
Transforming ideas into intelligent products through engineering, AI, and design. Specializing in production-grade backend scaling, vector search architectures, and polished interactions.
Identity Profile
I build sovereign AI engines and high-end interactive interfaces.
Computer Science and Technology undergraduate with hands-on experience building and shipping production-grade features using Python/FastAPI, React/TypeScript, and REST microservices on Azure and AWS. Proven track record as a Generative AI Developer and UI/UX Designer, delivering scalable backend APIs, intelligent RAG pipelines, CI/CD integrations, and component-driven frontends with polished user interactions. Hands-on experience with Docker, Git workflows, and system design paradigms.
dev & design
Click to rotate panel card
Structuring decoupled backends (FastAPI/Node) containerized on isolated volumes, paired with cursor coordinate highlights and custom cubic easings.
Click a bar to view achievements
Technical Stack & Skills
An interactive index of my programming languages, architectures, AI pipelines, databases, and deployment platforms.
Engineering localized retrieval systems (RAG), high-dimensional vector databases (FAISS), and relational graph networks (Neo4j) to build air-gapped intelligent agents.
AI & Machine Learning
Engineering localized retrieval systems (RAG), high-dimensional vector databases (FAISS), and relational graph networks (Neo4j) to build air-gapped intelligent agents.
Programming Languages
Backend Architecture
Frontend Engineering
Databases & Storage
Cloud & DevOps
SecOps & OWASP Defense
Gemini AI
Interfacing with Google Gemini models for multimodal generation and prompt workflows.
Kiro
Leveraging custom agent logic for complex, autonomous reasoning flows.
Antigravity
Utilizing advanced developer assistant environments for rapid context exploration.
Cursor
Using AI-assisted developer IDE layouts to accelerate codebase refactoring loops.
Claude Code
Executing automated terminal code audit checks and quick hotfixes.
Ollama (LLaMA 3)
Hosting completely local, air-gapped language models to avoid exposing sensitive workspace queries.
RAG pipelines
Structuring document ingestion, chunk overlap indexing, and semantic search context assembly.
scikit-learn
Applying mathematical algorithms for data clustering, regressions, and metric tracking.
pandas
Transforming tabular data, parsing dirty data sheets, and doing mathematical aggregations.
PyTorch
Building, training, and running forward/backward passes for machine learning tensor models.
Education & Coursework
Detailed academic records, specialized focus modules, and ongoing university credentials.
Dayananda Sagar University
Bengaluru, India
Bachelor of Technology - Computer Science and Technology
Ongoing CST Specialization — 2023–Present
System Capability Matrix
Visual proof of autonomous code engineering, infrastructure stability, and enterprise-grade security protocols.
Knowledge Graph & Vector Systems
Architecting air-gapped retrieval agents. Combining dense matrix searches in FAISS with rigid relational networks in Neo4j to mitigate Large Language Model (LLM) hallucinations.
from langchain_community.vectorstores import FAISS
from langchain_community.graphs import Neo4jGraph
class AirGappedRAGPipeline:
def __init__(self, index_path, neo4j_uri):
self.vector_store = FAISS.load_local(index_path, embeddings)
self.knowledge_graph = Neo4jGraph(url=neo4j_uri, auth=(user, pwd))
def query(self, prompt: str, rbac_role: str):
# MMR search matches context cleanly, passing to LLaMA 3
ctx = self.vector_store.max_marginal_relevance_search(prompt, k=4)
return llm.invoke(ctx, role=rbac_role)Microservices & Scale
Asynchronous FastAPI applications paired with enterprise Node.js clusters, operating with clean architecture principles.
SecOps & Verification
Proactive enforcement of OWASP Top 10 compliance: JWT token validation, rate-limiting, and validation schemas.
Cloud-Native Continuous Delivery
Containerizing microservices inside isolated Docker volumes deployed over AWS and Azure compute nodes. Controlled automated delivery cycle compressed release steps by 70%.
projects portfolio
dineingo
Smart Full-Stack Restaurant & Event Booking Platform
nexora
Sovereign Hybrid RAG for Air-Gapped Aerospace Intelligence
Click to explorelifeflow
Microsoft Imagine Cup 2026 Innovation
Click to explorerunagenai
Google Cloud Gen AI Exchange Hackathon 2025
Click to explorespitchai
JARVIS-Level Local AI Desktop Assistant
Click to exploremuseverse
AI Museum Curator (Kaggle ADK Capstone)
Click to exploreamazonml
Product Price Prediction (Machine Learning Solution)
Click to explorecyberconstituent-slm
Constitutional AI-Aligned Cybersecurity Threat Classifier
Click to explorelumaforge
AI Image Generation & Styling Platform (Optimized for Apple Silicon)
Click to exploreLumaForge
Introducing LumaForge – An Open-Source AI Image Generation Model
Ollama Integration
Local Ollama models intelligently expand prompts before diffusing, avoiding cloud APIs.
Hardware Acceleration
Optimised inference using FP16, PyTorch, and MPS for Apple Silicon, plus CUDA and TensorRT support for Windows.
Vectorised NumPy
Background removal vectorized in NumPy (running in ~8.9ms) and sketch filtration (running in ~4.1ms).
// THE CHALLENGE
Pushing local AI image generation limits without relying on expensive, proprietary cloud APIs.
I wanted to explore how far I could push local AI image generation without relying on cloud APIs. That curiosity led me to build LumaForge, an open-source image generation model built on Stable Diffusion 3.5 Medium and optimised for efficient local inference.
When I started this project, my initial idea was to build an ML-based prompt classification model that could identify image attributes and quality labels before passing prompts to the diffusion pipeline. However, building a reliable classifier required a much larger labelled dataset than I had available.
// THE RETROSPECTIVE & SOLUTION
Building a local-first workflow utilizing open models, hardware optimization, and vectorized post-processing.
Instead of stopping there, I took a different approach. Rather than relying on hosted LLM APIs such as OpenAI or Gemini, I built a local-first workflow. I integrated Ollama to intelligently expand user prompts into richer, more descriptive queries before sending them to the Diffusers pipeline. I used Ollama models because they are open models, unlike frontier proprietary models. This significantly improved prompt quality, composition, lighting, and overall adherence while keeping the workflow completely local.
Building LumaForge gave me hands-on experience with Stable Diffusion 3.5 Medium, Hugging Face Diffusers, prompt engineering, hardware acceleration (FP16, MPS, CUDA), and designing efficient local AI inference pipelines. More importantly, it taught me that sometimes changing the approach is better than forcing the original idea.
Core Focus Areas
NEXORA Graph-NLI Verification Visualizer
Click on any pipeline node to inspect real-time Graph-NLI verification stages, Cypher database records, and verification algorithms.
LLM Draft Response Generation
The base aerospace LLM generates an unverified draft response containing potential factual hallucinations.
# Base LLM response generation
draft_response = base_model.generate(
prompt="Describe the cooling valve component of the F16 wing.",
max_tokens=256
)
# Output contains unverified structural claims
print(f"Draft: {draft_response.text}")Design Systems Showcase
Interactive canvas presenting high-fidelity custom frontend wireframes, hover micro-interactions, and underlying CSS design tokens.
Hover Mouse Coordinate Tilt
Move your cursor over this card to witness three-dimensional tilt angle matrices.
/* Retro 8-Bit Design Tokens */
:root {
--color-bg-light: #F7F7F5;
--color-primary: #111111;
--color-secondary: #555555;
--color-accent: #C7FF3D;
--color-success: #4ADE80;
--color-border: #111111;
--shadow-offset: 4px;
}Academic Credibility
Reviewing deterministic answers in air-gapped, zero-network environments.
NEXORA: Sovereign Hybrid Retrieval-Augmented Generation for Air-Gapped Aerospace Mission Intelligence
ABSTRACT
This paper presents NEXORA, a sovereign offline Retrieval-Augmented Generation (RAG) framework optimized for air-gapped aerospace intelligence environments. Because public cloud APIs are blocked due to security regulations, NEXORA uses local embeddings and private models. To address the problem of Large Language Model (LLM) hallucinations, we join dense mathematical vector stores (FAISS) with structured knowledge network layers (Neo4j graph schemas). This hybrid pipeline ensures deterministic answers, validates user access clearance levels (RBAC), and handles complex relationship paths with sub-second retrieval times, outperforming traditional semantic similarity architectures.
1. INTRODUCTION
In safety-critical domains such as aerospace engineering, information queries must return factual, deterministic results. Traditional RAG systems query flat vector embeddings, which lack structural entity-relationship maps. This paper introduces a hybrid model where local dense lookups trigger Cypher traversal queries to reconstruct complex entities.
2. ARCHITECTURE
Our system separates unstructured manuals from structural telemetry maps. FAISS handles semantic search queries to extract candidates, while a parallel path Traversal routine inspects connections in Neo4j. The contexts are merged in a validation container before LLaMA 3 executes responses.
Production Chronicle
The chronological mapping of my academic foundations, hackathons, and systems integration work.
B.Tech Computer Science & Technology
Matriculated into Computer Science and Technology B.Tech at Dayananda Sagar University. Mastery of Core Foundations: Algorithms, Data Engineering, and System Design.
NASA Space Apps Arena & Hackathons
Entered NASA Space Apps Challenge global hacking arenas, pioneering rapid prototyping methods.
Enterprise Deployments & ERP Backends
Produced institutional scale products (FlowGrid ERP backend engineering, DineInGo baseline conceptualization) and acquired enterprise systems credentials (AWS Cloud Foundations, Google Cloud Automated Delivery systems).
Advanced Engineering Milestones
Advanced Engineering Milestones. Participated in Microsoft Imagine Cup, completed complex RAG architectural research papers, and finalized full-stack microservices platforms.
Credentials Hover Board
Clickable or hoverable verification panels matching certified capabilities.
NASA Space Apps Challenge
Global Participant (Consecutive years: 2024 & 2025)
Let's Build Something Meaningful.
Have an enterprise backend scale requirement, a vector index pipeline to design, or a complex React interactive dashboard to compile? Let's align.