2023 - Present
Bachelor Degree
Indian Institute of Technology Jodhpur (IITJ)
Hello, I'm
K V Raghava Karthikeya
I'm an aspiring AI ML & Computer Vision Engineer and Web & Mobile Developer.
constcoder={name:'K V Raghava Karthikeya',skills:['C/C++', 'DSA', 'Kubernetes', 'API Integration', 'React', 'HTML & CSS', 'Python', 'n8n', 'OpenCV', 'PyTorch', 'Tensorflow', 'DeepLearning', 'Keras', 'Express', 'NodeJS', 'MySql', 'MongoDB', 'Docker', 'AWS'],hardWorker:true,quickLearner:true,problemSolver:true,hireable:function() {return(this.hardWorker&&this.problemSolver&&this.skills.length>=5);};};About Me
I am an aspiring AI/ML Engineer and Full-Stack Developer at IIT Jodhpur, specializing in intelligent systems that bridge complex data and real-world impact. I am actively building agentic workflows, computer vision models, and highly scalable web architectures.
Currently Working On
Local AI Agent System — Full Documentation & Portfolio Update
2026-06-22Update! : Successfully documented and published the complete architecture of the Graph-Augmented Multi-Agent Orchestrator. The entire system runs 100% locally via LM Studio. Built a fully custom Graph RAG from scratch — SQLite , recursive CTE lineage traversal, cross-session linking, and token-budget-aware context assembly. Engineered Progressive Depth Escalation: a query-aware retrieval system where greetings skip history entirely while deep research questions trigger full BFS graph walks across sessions and summaries — three-tier fallback. Designed a hierarchical Level Summarizer that compresses each graph depth into summary paragraphs after every request, creating a memory pyramid stored in both SQLite and ChromaDB. Deployed 4 specialist agents (news, mail, calendar, task) with a manager-worker LLM architecture. Integrated this with WhatsApp with its own independent graph, entity index, and vector collections.
Project Objective
Local AI agent built from scratch. Contains multi agentic architecture with planner/worker/feedback models with adaptive context scaling.
Multi agent system - Metrics & Demo
2026-06-14Update! : Final metrics published: average inference latency 320ms, token usage optimized to 1.2K per simple query, tool calling success rate 94%. Video demo and GitHub code release pending.
Project Objective
Local AI agent built from scratch. Contains multi agentic architecture with planner/worker/feedback models with adaptive context scaling.
Experiences
Python Automation Intern (AI, n8n, Web Scraping)
Figmenta •
- Engineered 15+ production-grade automation workflows handling 6,000+ data points per run for large-scale social platform data processing.
- Designed multi-stage ETL pipelines that scraped platform data, stored structured outputs, and triggered API-driven downstream workflows.
- Integrated database-backed pipelines with AI agents to compute Z-score-based analytics, turning raw data into actionable marketing signals.
- Built dynamic workflow chains to provision R2 storage buckets and serve structured media outputs.
PROJECTS
AI/ML
Local AI Agent System
A fully local, security-first multi-agent system with a custom Graph RAG memory architecture — built entirely from scratch on SQLite (13 tables, recursive CTEs), ChromaDB (4 collections), and local LLMs via LM Studio. Features Progressive Depth Escalation: per-query retrieval depth (0–3) where greetings use a shallow SQLite lookup and deep research questions trigger a full BFS graph walk across sessions, entities, and summaries. Includes deterministic regex-based entity extraction (microsecond-level, no embeddings), hierarchical level summarization creating a memory pyramid, and 4 specialist agents (news, mail, calendar, task) with manager-worker LLM separation (Qwen 3.5 9B/2B). All inference runs 100% locally — zero cloud dependency.
PythonFastAPIUvicornPydantic+8 more
Cancer-Detection-Model
A machine learning–based cancer detection model designed to analyze histopathology images to identify patterns indicative of malignant disease. The project compares two CNN architectures: a Baseline model (~4M parameters) with three dense layers and a smaller Advanced model (~1.5M parameters) using Global Average Pooling and dropout regularization. Despite having 2.6x fewer parameters, the Advanced model achieves higher validation accuracy (92% vs 83%) while being more memory-efficient and less prone to overfitting. The model leverages supervised learning to distinguish between cancerous and non-cancerous cases, enabling early detection, risk stratification, and decision support for clinicians.
PythonPyTorchtorchvisionOpenCV+7 more
Gesture-Based Mouse Control
A real-time touchless human-computer interaction system that uses a webcam to capture hand gestures and control mouse functions. The system recognizes a 5-gesture vocabulary — Cursor Control (open hand with index finger extended), Left Click (index-middle finger tap), Right Click (middle-ring finger tap), Drag & Drop (closed fist with movement), and Mission Control (all five fingers spread). A calibration system uses adaptive skin detection and background subtraction to work across different lighting conditions and skin tones. Achieves 94.7% recognition accuracy with 95-100ms end-to-end latency using a custom 5D nearest-neighbor classifier on MediaPipe hand landmarks.
PythonOpenCVMediaPipeNumPy+2 more
3D Motion Tracking System
End-to-end gesture recognition pipeline for Arduino Nano 33 BLE Sense using TensorFlow Lite Micro. Captures 6-axis IMU data at 100Hz, trains a Keras neural network (714→50→15→4), and deploys quantized TFLite inference in under 2ms on an ARM Cortex-M4 microcontroller with 92% accuracy across 4 gesture classes.
PythonTensorFlowKerasTensorFlow Lite Micro+5 more
MiPa — Mixed Patch Object Detection
A research critique and experimental reproduction of the WACV 2025 paper 'Mixed Patch Visible-Infrared Modality Agnostic Object Detection.' Implemented and compared 5 mixing-ratio distributions (Uniform, Beta(2,2), Beta(0.5,0.5), Truncated Uniform, Gaussian) across CNN (ResNet50-FPN) and ViT (Swin-Tiny) backbones for pedestrian detection on the LLVIP dataset. Key finding: Beta(2,2) consistently outperforms the paper's default Uniform(0,1) distribution — achieving +1.24% RGB AP50 improvement on CNN backbones. Note: Due to Kaggle GPU session constraints, CNN backbone results were partially saved; the ViT backbone experiments have complete result logs with consistent improvement across all mixing distributions.
PythonPyTorchtorchvisionFaster R-CNN+4 more
MedTech — Cancer QoL Prediction
A machine learning project predicting Quality of Life (SF-36 scores) in Head & Neck Cancer patients by combining clinical/demographic data with Ayurvedic Prakriti (Vata/Pitta/Kapha) body constitution assessment. Engineered 142 features from 226 patient records including TNM staging, cancer site encoding, and 36 SF-36 quality-of-life indicators. Achieved R² = 0.88 using XGBoost regression with RandomForest baseline.
PythonXGBoostscikit-learnPandas+3 more
Web Dev
Personal Portfolio
A modern, high-performance developer portfolio built with Next.js 13+ App Router and Tailwind CSS. Features a responsive dark-themed design with glassmorphism effects, dynamic project showcases with letterboxed previews, a live project timeline tracking ongoing development, and integrated contact form. Deployed on Cloudflare Pages with Google Tag Manager analytics.
Next.jsTailwind CSSReactSCSS+1 more
Route Optimisation (Google Maps Clone)
A real-time route optimization algorithm integrated into a web-based application that computes the shortest and most efficient paths using live data. The system dynamically processes factors such as distance, travel time, and routing constraints to deliver optimal navigation results. Paired with a user-friendly interface and interactive map features comparable to Google Maps, it enables seamless route visualization, turn-by-turn guidance, and responsive updates to enhance usability and decision-making.
PythonC++JavaScriptFlask+6 more
Manhwa Agent Bot
An automated offline-first dashboard that tracks personalized reading lists from Asura Scans. Uses a Playwright-based browser emulation layer to detect new unread chapters and displays them in a Flet desktop widget. An n8n workflow orchestrates 6-hourly scans with Docker containerization, automatically opening new chapters in the browser.
PythonFastAPIPlaywrightBeautifulSoup+5 more
Spotify Widget for macOS
A learning-focused macOS widget that integrates with the Spotify Web API via OAuth 2.0 to display and control currently playing music. Built as a utility project to explore OAuth Authorization Code Flow, automatic token refresh mechanisms, and REST API integration patterns. The widget retrieves real-time playback state (track title, artist, album art, progress), enables play/pause/skip controls, and displays recently played tracks. Note: This is a learning/utility project exploring Spotify API integration patterns — not a production-grade macOS widget.
PythonFlaskRequestsSpotify Web API+1 more
Panda Sleep Studio Portfolio
A polished single-page portfolio website for 'Panda Sleep Studio' — a fictional creative agency blending sleep culture with design technology. Built with Vite + React 18 featuring GSAP scroll-triggered animations, custom lerp-based cursor system, Lenis smooth scrolling, video-on-hover project cards, CSS marquee banners, and full-screen slide-down drawers. Warm beige aesthetic with hand-written CSS (no frameworks).
JavaScriptReactViteGSAP+3 more
Skills
Educations
2021 - 2023
Higher Secondary Certificate
Narayana Junior College
till 2021
Secondary School Certificate
DAV Public School