Keerthi K

AI/ML Engineer | Software Developer

Passionate about writing code that matters — blending functionality, scalability, and intelligence.

Contributor to open-source projects and constantly learning to push the limits of what I can build.

PythonReact & Node.jsFlutterLLMsML Models

Development

Python, C++, Java
React, Node.js
FastAPI, Flask
SQL, NoSQL
Docker, Redis

AI / ML

Scikit-learn, NumPy
Transformers (HF)
Vector Search (FAISS)
RAG & Multimodal
Prompt Engineering

Tools & Infra

Git, GitHub/GitLab
Firebase, GCP, AWS
Postman, Swagger
Jira, Notion
Docker, Redis

Education

New Horizon Public School

ICSE Board – 10th Grade

2006 – 2020

Mount Carmel PU College

PCME (Physics, Chemistry, Math, Electronics)

2020 - 2022

PES University

Computer Science and Engineering

2022 – 2026

Highlighted Technical Projects

FinSage – AI-Powered Finance Assistant

A full-stack personal finance platform offering smart insights, predictive budgeting, expense categorization, goal tracking, receipt OCR, and a natural language chatbot.

User Experience

• Real-time dashboard with category-wise spending
• Responsive UI using React & Tailwind
• OAuth 2.0 login with Gmail integration
• Visual analytics via Recharts

Tech Stack & Intelligence

• FastAPI backend with PostgreSQL
• Modular Dockerized deployment
• LLM-powered expense categorization
• Chat assistant for query-based insights
• Forecasting via time series models

Highlights

• Gmail parser for auto-logging transactions
• Tesseract OCR for receipt-based logging
• AI queries like “What did I spend on food last week?”

System Architecture

Multimodal PDF Q&A System

A scalable RAG system combining text and image understanding for accurate document QA, using Sentence-BERT, CLIP, FAISS, and LLaMA with an agentic reasoning pipeline.

Core Capabilities

• Multimodal Embedding via SBERT & CLIP
• Hybrid Retrieval (FAISS + BM25)
• Agentic Workflow & Planner Agent
• Web Search Agent (Tavily API)

Model & Fine-tuning

• LLaMA 3.2B for response generation
• LoRA fine-tuning on HotpotQA
• Contextual late fusion of image/text
• Memory Agent for session tracking

Key Outcomes

• Reduced LLaMA 3.2B to 1.2B boosting speed for real-time QA
• LoRA boosted EM to 8% and F1 to 79%
• Modular architecture, production-ready

System Architecture

Yet Another Task Distributor

A fault-tolerant distributed task queue for emergency services, ensuring exactly-once execution and intelligent routing. Built with Python, Kafka, and Redis, it supports dynamic worker orchestration, heartbeat monitoring, and real-time status tracking.

Client Features

• Unique Task ID & Redis-backed status registry
• Real-time result querying
• Emergency call simulator for load testing

Worker Features

• Kafka consumer groups for load balancing
• Heartbeat monitoring & smart retries
• Structured status updates

System Highlights

• Exactly-once execution guarantee
• Redis-based low-latency result caching
• Autoscaling with dynamic load handling

Architecture Overview

Smart Diet & Fitness Planner

A cross-platform app (Flutter + FastAPI) that offers AI-driven meal recommendations based on activity, intake, and biometrics. Uses Firebase Auth, Google Calendar, and on-device ML for private, low-effort tracking backed by an Indian/global nutrition DB.