Nithees Kanna | AI & Machine Learning Developer

Hey Hi! I am NK

I am your friendly neighbourhood NAAH I AM VENGEANCE!!!

General Overview

About Nithees Kanna

I am a Python Machine Learning and AI Developer focused on building robust retrieval pipelines, agentic RAG workflows, vector search layers, and automated MLOps pipelines. I specialize in deploying fine-tuned models onto Google Vertex AI and orchestrating multi-agent systems that solve complex, real-world data challenges.

Milestones

Professional Journey

2020 — 2024

Bachelor of Information Technology

St. Joseph's College of Technology

Completed a Bachelor of Information Technology at St. Joseph's College of Technology, building strong foundations in programming, databases, and machine learning.

2022

Remote Intern

Bolt IO

Remote internship at Bolt IO — gained hands-on industry experience working on real-world software and data tasks alongside coursework.

2024

Incubation Training in AI Systems

Kanini

Completed Kanini's incubation training in AI systems, working hands-on with machine learning pipelines and applied AI solution building.

Present

Perpetuitti Technosoft

Current Role

Currently at Perpetuitti Technosoft, building AI and machine learning solutions — from agentic workflows to production ML systems.

Stack

Technology & Frameworks

Selected Positioning

What I can do in an AI team

01 AI Agents

AI Agents and agentic workflows

Build retrieval pipelines, agentic RAG flows, multi-agent setups, vector search layers, and evaluation-ready AI application logic.

02 Model Engineering

Model Training, Development, Evaluation, Deployment and Monitoring

Work across data collection, cleaning, model training, validation, evaluation, deployment, and continuous production monitoring.

03 MLOps

MLOps - Google Cloud Platform, Kubernetes and Docker

Automate containerized deployment pipelines, manage container orchestration using Kubernetes, and deploy robust architectures on Google Cloud Platform.

Selected Deployments

Featured Projects

Project Lovelace

An autonomous reasoning AI Agent for STEM fields designed to solve complex multi-step math proving, scientific simulation, and symbolic calculation tasks with self-correcting loop logic.

Engineered custom reflection prompts yielding a 35% accuracy improvement in multi-step STEM proofs.
Configured interactive Python sandboxes and WolframAlpha tool bindings.
Designed self-correcting reasoning loops to validate outputs dynamically.

AI Agents LangGraph STEM Reasoning Python SymPy

02

Orchestra-RAG

A multi-agent RAG framework using LangGraph to dynamically plan, retrieve, and validate queries across multiple documentation layers.

Achieved 98.4% retrieval accuracy via custom reranking layers.
Implemented self-correcting retrieval loops using LangGraph.

Python LangGraph Pinecone RAG

03

AI Workflow Identifier

A multi-agent AI workflow that analyzes uploaded documents to identify semantics — statements, syntax, and expected outputs — with specialized agents coordinating each stage.

Orchestrated multiple AI agents to parse, classify, and cross-validate document semantics.
Automated statement, syntax, and output identification for uploaded documents.

AI Agents Python LangChain NLP

04

Documentation Bot for Legalised Works

An agentic documentation assistant that drafts, reviews, and organizes legalised documents using cooperating AI agents for clause structuring and validation.

Deployed cooperating AI agents to draft and cross-review legalised documentation.
Reduced manual drafting effort with automated clause structuring and validation.

AI Agents LLM Document Automation Python

Core Work & Research

Academic & Applied ML

05

File Security Analyzer

A security scanning application checking local files for weak protections and permission configurations using ML techniques.

Built classification layers using TensorFlow and Dense neural networks to predict vulnerability scores.
Analyzed access control logs to identify risky user privilege delegations.

TensorFlow Security Neural Networks Python

06

Wastewater Quality Classifier

An environmental monitoring system classifying water safety and identifying potable thresholds in wastewater sumps.

Integrated sensor data streams measuring physical and chemical water properties.
Implemented a Convolutional Neural Network (CNN) in Python to categorize safety parameters.

Python CNN Sensors Data Analysis

07

License Plate Detector

A deep-learning computer vision system automating license plate detection and reading from vehicle images.

Developed a Convolutional Neural Network (CNN) model for bounding box region detection.
Applied deep regression networks to locate and isolate license text under varied light levels.

CNN Deep Regression Computer Vision TensorFlow

08

Spotify Music Segmenter

A customer segmentation application categorizing listeners based on music taste and database profiles.

Analyzed user listen histories, tracking acoustic qualities across different genres.
Implemented clustering models to identify distinct listener persona buckets.

Clustering Spotify API Data Mining Python

09

Face Mask Detector

A real-time vision classifier detecting face masks in video streams and logging compliance data.

Built detection and tracking scripts using OpenCV and CNN models in TensorFlow.
Integrated Pytesseract for reading text and generating log entries for monitoring feeds.

OpenCV TensorFlow Pytesseract CNN

10

Gesture Media Controller

A human-computer interaction system translating real-time hand movements into PC media controls.

Trained lightweight gesture mapping networks using TensorFlow.
Used OpenCV streaming to trigger play, pause, and volume changes hands-free.

TensorFlow OpenCV Computer Vision Python

Capabilities

Core Technical Matrix

Select any capability above to see real-world details, key tools, and project impact.

How I Work

From problem framing to maintainable AI

1

Frame the problem

Clarify objective, data shape, metrics, constraints, and the expected user workflow.

Activities: Stakeholder alignment, metric definition (e.g., NDCG, recall), and data profiling.
Deliverables: Baseline evaluation sets, system architecture diagrams, and scoping documentation.

2

Build the system path

Connect data, retrieval, model logic, APIs, and automation into one testable flow.

Activities: End-to-end pipeline prototyping, LLM orchestration (LangGraph), and tool integration.
Deliverables: Working FastAPI microservices, relational/vector schemas, and mock-tested code paths.

3

Measure and harden

Validate outputs, add checks, monitor behavior, and prepare for iteration in production.

Activities: Latency and accuracy profiling, safety guardrail setup, and CI/CD pipelines.
Deliverables: LLM monitoring dashboards, security integration (input/output filters), and deployment setups.

Continuous Learning