Skip to content
View kokilamariyayi's full-sized avatar
7 followers · 1 following

Achievements

Achievement: Quickdraw

Achievements

Achievement: Quickdraw

Block or report kokilamariyayi

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
kokilamariyayi/README.md

Kokila M

Typing SVG

Academic Location Graduation




Profile Views Followers Stars

About Me

I'm a final-year B.Tech Artificial Intelligence & Data Science student at Nehru Institute of Engineering and Technology, Anna University, with a strong focus on GenAI/LLM systems, agentic architectures, and applied deep learning. My engineering approach centers on shipping production-shaped systems — RAG pipelines, self-correcting agents, and evaluation frameworks — validated with real benchmarks rather than toy demos.

I care about the full stack of a model-backed product: retrieval, orchestration, evaluation, and the UI that ships it. Recent work spans multilingual RAG chatbots, LLM red-teaming at scale, agentic SQL generation, and computer vision pipelines deployed end-to-end on Firebase, FastAPI, and Azure.

Open To: AI/ML Engineer • GenAI/LLM Engineer • Applied AI roles @ product companies & MNCs

Tech Stack

Only what I've actually shipped with, across real projects.

Languages

Python Java SQL

AI / ML / GenAI — core focus across every project

TensorFlow Keras scikit-learn LangChain LangGraph Groq YOLOv8

Cloud & Tooling — deployment for VidyaPath, Parsify AI, vehicle-damage-assess

Azure Firebase Hosting Git GitHub Actions

AI / ML Expertise

Domain Proficiency Details
RAG & Agentic Systems ★★★★★ LangChain, LangGraph, ChromaDB, self-correction loops, multi-agent orchestration
LLM Evaluation & Red-Teaming ★★★★★ Adversarial prompt design, failure taxonomy, MART-style iterative hardening
Computer Vision ★★★★☆ Transfer learning (VGG16, EfficientNet, MobileNetV2, YOLOv8)
Classical ML ★★★★☆ Ridge/Lasso/Random Forest, fraud detection, sentiment models
LLM APIs & Inference ★★★★★ Groq (Llama 3.3 70B, Whisper), OpenAI-compatible tooling
Full-Stack Deployment ★★★★☆ FastAPI, Flask, React, Firebase Hosting, Azure

Featured Projects

🔹 Adaptive Text-to-SQL Agent

Self-correcting natural-language-to-SQL agent that iteratively refines failed queries using execution feedback.

Stack Scale Performance Security Impact Repository
LangGraph, LangChain, Groq, ChromaDB, FastAPI Multi-turn query correction loop 75% → 91.7% accuracy with self-correction Parameterized query execution Production-shaped agentic SQL pipeline text-to-query
🔹 LLM Evaluation & Red-Teaming Framework

A structured adversarial evaluation suite for LLM robustness testing across multiple failure modes.

Stack Scale Performance Security Impact Repository
Python, Groq, custom eval harness 520+ adversarial prompts, 6 failure categories 17.6% violation-rate reduction via MART loop Full git-history secret remediation Reusable red-teaming framework for LLM apps llm-eval-redteam
🔹 VidyaPath — Multilingual RAG Career Guidance Chatbot

Multilingual RAG chatbot guiding Grade 8–12 Indian students through career decisions, deployed live.

Stack Scale Performance Security Impact Repository
LangChain, ChromaDB, Groq Llama 3.3 70B, FastAPI, Firebase Multilingual (regional Indian languages) Low-latency browser-direct Groq inference Prompt-injection hardened Live student-facing deployment vidyapath
🔹 Self-Correcting LLM Agent Pipeline

Agentic pipeline benchmarked across diverse queries with iterative self-correction.

Stack Scale Performance Security Impact Repository
LangGraph, Groq Benchmarked on 30+ queries 76.7% success rate Sandboxed execution Demonstrates agentic reliability patterns self-correcting-llm-agent
🔹 SpendGenie — Expense Forecasting Tool

Personal finance forecasting tool with real regression benchmarking, not a black-box demo.

Stack Scale Performance Security Impact Repository
Flask, React, scikit-learn Ridge, Lasso, Random Forest benchmarked 48.3% MAE improvement over baseline Local data handling Genuine model comparison, not a toy forecast expense-forecasting-ai-tool
🔹 Vehicle Damage Assessment System

End-to-end vehicle damage detection and cost estimation pipeline, deployed to Azure.

Stack Scale Performance Security Impact Repository
YOLOv8, EfficientNet-B4, Azure Damage detection + PDF report generation Automated cost estimation pipeline Cloud-hosted inference Deployed insurance-style assessment tool vehicle-damage-assess
🔹 Parsify AI — Document Extraction App

Document classification and field-extraction app with voice transcription support.

Stack Scale Performance Security Impact Repository
Groq, FastAPI, React, Firebase Multi-document-type classification Groq Whisper transcription integrated Firebase-hosted auth Live deployed document intelligence tool parsify-ai

Experience

Agentic AI Intern — Nxtlogic Software Solutions Dec 2025 – Jan 2026

Worked on agentic AI system design and orchestration patterns for production-shaped LLM workflows.

  • Designed and tested multi-step agentic pipelines
  • Contributed to evaluation and reliability tooling

LangGraph Agentic AI LLM Orchestration

AI Intern — Infosys Springboard 6.0 Sep 2025 – Dec 2025

Applied AI/ML engineering across structured project milestones under the Infosys Springboard program.

  • Built and evaluated ML/DL models against defined project checkpoints
  • Delivered structured documentation and technical reporting

Python Machine Learning Deep Learning

AI & Cloud Intern — Edunet Foundation / AICTE Jul 2025 – Aug 2025

Extended a vehicle damage assessment computer vision system with cloud deployment.

  • Deployed YOLOv8 + EfficientNet-B4 damage detection pipeline to Azure
  • Built automated cost estimation and PDF reporting

YOLOv8 EfficientNet Azure

Data Analytics Intern — NoviTech R&D Aug 2024 – Sep 2024

Focused on data analytics workflows and reporting for R&D use cases.

  • Performed data cleaning, analysis, and visualization
  • Delivered analytical insights to support R&D decisions

Data Analytics Python Visualization

Achievements

Recognition Details
Microsoft Learn Student Ambassador Community Influencer Path, ambassador ID studentamb_506764
Rotaract Club of NIET — Young Leader Chaired events including Kandangi (Pongal-themed event)
LeetCode Consistency 142+ active days toward 150-day streak badge
NPTEL Discipline Star & Enthusiast Star 8+ NPTEL courses completed

Certifications

AWS AWS

Google Google AI/ML

IBM IBM GenAI

Microsoft Microsoft Learn

NPTEL NPTEL

Coding Profiles

LeetCode GeeksforGeeks

Contribution Activity

Current Focus

Learning:
  - Advanced agentic orchestration patterns
  - PEFT / LoRA / QLoRA fine-tuning strategies
  - LLM evaluation & alignment methods

Building:
  - LLMs Unfiltered — LinkedIn newsletter on foundational NLP
  - Portfolio site redesign (Token Stream design system)
  - VGG16 transfer-learning image classification pipeline

Exploring:
  - Multi-agent systems
  - Production-grade RAG evaluation

Open To:
  - AI/ML Engineer roles
  - GenAI / LLM Engineer roles
  - Product-focused engineering teams

Connect

"Beyond the prompt, inside the model."

Footer

Pinned Loading

  1. llm-eval-redteam llm-eval-redteam Public

    Automated LLM red-teaming framework testing 520+ adversarial prompts across 6 failure categories using DeepEval, LangSmith & Groq

    Python 2

  2. resume-roaster resume-roaster Public

    AI-powered resume analyzer that gives brutally honest, LLM-generated feedback on resumes built with Python and NLP to score and critique like a recruiter would.

    Python 5

  3. linkedin-post-generator linkedin-post-generator Public

    AI-powered LinkedIn post generator built with Groq (Llama 3 70B) and Streamlit. Supports 6 post types, 4 tone modes, and generates up to 3 unique versions per topic. Deployed as a live web app on F…

    CSS 5

  4. expense-forecasting-ai-tool expense-forecasting-ai-tool Public

    AI-powered expense prediction tool that analyzes historical financial data to forecast future spending trends with optimized regression models.

    JavaScript 5

  5. self-correcting-llm-agent self-correcting-llm-agent Public

    Self-correcting LLM agent with a built-in reflection loop detects flawed reasoning mid-run, retries up to 3x, and gracefully falls back instead of hallucinating. Built with LangGraph + Groq (llama-…

    Python 2

  6. text-to-query text-to-query Public

    Agentic Text-to-SQL pipeline with a self-correcting execute→error→retry loop (LangGraph) and ChromaDB schema retrieval - improved execution accuracy from 75% to 91.7% across a 12-query benchmark.

    Python 2