const developer =
Adhithya Srivatsan
// Welcome to my terminal portfolio. // Type commands below or click links to navigate.
01. About Me
> Hello! I'm Adhithya Srivatsan, currently pursuing B.Tech in AI and DS at Shiv Nadar University.
> I've been captivated by coding since 9th grade, leading me to dive into web development and nurture aspirations of becoming an AI Engineer.
> I've built this website to showcase my coding journey. Eager to embrace every coding challenge, I invite you to explore my projects, and don't forget to check out the code on my GitHub — the links are under my pinned projects.
Happy coding exploration!
{
"name": "Adhithya Srivatsan",
"education": "B.Tech AI & DS",
"university": "Shiv Nadar University",
"interests": [
"Web Development",
"System Design",
"Agentic AI",
"Data Science",
"Machine Learning/Deep Learning"
],
"status": "Coding..."
}02. Experience
AI Intern @ Indium Software
Remote
Built an agentic API security testing platform over an 8-week internship. Instead of static test scripts, the system reads an API spec, reasons about it, and writes its own security-focused test cases.
- >Designed a closed observe–analyze–regenerate loop in LangGraph with PostgreSQL-backed state checkpointing, so the system checks its own test coverage and generates more targeted tests in each reasoning cycle.
- >Built a two-pass HTTP execution engine that captures runtime authentication and setup context before running dependent test cases, removing the need to hardcode an auth flow for every endpoint.
- >Integrated Gemini models through LiteLLM with few-shot prompt libraries to generate structured test cases across OWASP API security categories: BOLA, authentication bypass, injection and mass assignment.
- >Deduplicated generated test payloads at the database level using SHA-256 canonicalization in PostgreSQL.
- >Set up a GitHub Actions CI pipeline that lints, type-checks and runs tests on every push and pull request.
03. System Diagnostics
04. Projects
Replaces the email-based negotiation between capstone teams and review panels with a small booking system: panels publish slots, teams request them, the panel votes, and the outcome is visible to everyone instead of buried in an inbox.
Give it a video URL + a line of dialogue -> get back the exact timestamp frame number matched text, confidence, and the rendered frame image
Benchmarks how sentiment models trained on 48K general YouTube comments transfer to a single video, comparing text-cleaning variants, embeddings, classifiers and a fine-tuned RoBERTa.
EfficientNet-B5 detector for the NTIRE 2026 challenge that fuses RGB with an FFT frequency channel to spot AI-generated images under real-world distortions.
A customer care chatbot (agent) built using langchain with AWS Runtime in Bedrock Agentcore.
An intelligent agent capable of autonomously finding relevant news articles on a given topic and compiling them into a professional, engaging newsletter.
AI-driven communication system for seamless translation between speech, text, and Indian Sign Language (ISL).
AI-based travel planning tool that generates personalized 7-day itineraries based on user preferences using Crew AI, Streamlit, and GPT-4.
Allows users to upload documents and ask questions about their content using AI. Features secure auth and efficient document management.
Utilizes YOLOv8 for real-time object detection and the SORT algorithm for tracking to accurately detect and count vehicles passing through a predefined line in a video feed.
Computer vision project allowing users to adjust system volume using simple hand gestures by tracking thumb-index finger distance.
Deep learning model predicting customer churn for a bank using TensorFlow's ANN, trained on a dataset of 10,000 customers.
05. Achievements
Winner at GDG Hackatonic 2025
Secured 1st place out of 111 teams at PSG iTech, Coimbatore. Built GeoSense, a geospatial intelligence solution, during a 24-hour hackathon organized by Google Developer Groups.
Runner up at Smart India Hackathon 2024
Finalist at SIH Grand Finale held by Ministry of Education, GoI. Developed LINGUA, an ISL-to-speech and speech-to-ISL translation solution addressing diverse societal use cases. Secured runner-up position after a 36-hour development sprint.
06. Publications
Patch-Based Contrastive Learning with Latent Space Clustering for Unsupervised Sound Anomaly Detection
DCASE 2025 Challenge, Task 2 · Technical Report · 2025
Our submission to DCASE 2025 Task 2, which asks for first-shot detection of faulty machine sounds using only recordings of normal operation. We split log-Mel spectrograms into overlapping patches, encode each patch with a ResNet-34 trained contrastively, and pool them with attention that can also use machine metadata when it's available. Test clips are scored by Mahalanobis distance to normal clusters for each domain. The method beat the official baselines in target-domain AUC on 5 of 7 machine types and reached 94.16% source-domain AUC on valve.