Monish Saravanan

Master of AI & ML — Adelaide University

Monish
Saravanan

I'm training to become an AI architect who builds systems that solve problems people can't ignore.

Scopus-published 4 ML projects Computer vision Explainable AI

people have visited this portfolio

About me

I'm a computer science graduate from Chennai, now at Adelaide University completing my master's in Artificial Intelligence and Machine Learning.

My undergraduate degree at Sathyabama Institute of Science and Technology gave me a foundation in ML, computer vision, and NLP. Since then I've been drawn to problems where AI has to deal with things that are hard to see — translucent particles in water, hidden data in images, patterns in messy text.

Right now I'm focused on attention-based architectures and how neural networks can be made to selectively ignore irrelevant noise and focus on what actually matters in difficult, real-world conditions.

Education

2025 — Present
Master of Artificial Intelligence and Machine Learning Adelaide University, Australia
2021 — 2025
B.E. Computer Science — AI & ML Sathyabama Institute of Science and Technology, Chennai

Projects

Machine Learning · Explainable AI · Full Stack

FairHire AI — Explainable Candidate Ranking

Work in Progress · v0.9
Python FastAPI Next.js LightGBM SHAP pgvector Docker

A recruiter uploads a job description and a batch of resumes. The system parses them, ranks every candidate against the job, and shows the recruiter exactly why each person got the rank they did — with the actual lines from the resume that drove the score. Every ranked list goes through a human before any action is taken.

The pipeline runs across seven services: ingestion handles PDF and DOCX parsing with OCR fallback, an NLP layer extracts skills, experience, education and normalises them against the ESCO ontology, sentence-transformers embed both the job requirements and resume evidence into vectors stored in pgvector, and a LightGBM ranker scores each candidate. SHAP values tie every top feature back to a specific span of resume text — so the recruiter sees "ranked #1 because of 6 years Python, matches requirement 3" not just a number.

7 Backend services — ingestion, NLP, embeddings, ranker, explanations, fairness, API
9+ Features shipped — ranked list UI, candidate detail, fairness tab, PDF export, audit log
64 Synthetic eval pairs across 4 roles — end-to-end pipeline verified

Includes a fairness instrumentation tab that shows subgroup score distributions and top-K selection rates — with an explicit disclaimer that it is instrumentation for a real audit, not a bias removal claim. Hand-labelled evaluation set and live demo are the remaining milestones.

View source code, architecture docs, and system card GitHub →
Computer Vision · Deep Learning

EcoScan: Detecting Micro-Plastics in Waterways

Python YOLOv8 CBAM PyTorch OpenCV

Micro-plastics — particles smaller than 5mm — blend into sediment, surface reflections, and organic matter. They're translucent, tiny, and nearly impossible for a standard detection model to pick out in drone footage of a river.

I extended YOLOv8 with Convolutional Block Attention Modules (CBAM) inserted directly after the C2f backbone blocks, before the multi-scale fusion stage. This forces the network to suppress background noise — glare, ripples, colour variation — and amplify the faint signals left by translucent particles, rather than treating everything equally. Prior work only adds attention at the detection head; putting it earlier means the benefit carries through every downstream stage.

10,000+ Labeled training images
6 Debris classes (bottles, bags, foam, nets, other, micro-plastics)
4.9% Micro-plastic share — severe class imbalance addressed with focal loss

Training handled a sharp imbalance — micro-plastics are only 4.9% of annotations while macro-plastics account for 56%. I used focal loss to stop the model from ignoring them entirely, then oversampled micro-plastic examples with domain-specific augmentations: simulated murky water (brownish-green tint), Gaussian blur for camera defocus, and gamma correction to replicate changing light. YOLOv8's mosaic augmentation added further context diversity.

View source code and documentation GitHub →
Deep Learning · Security

Deep CNN Encoder-Decoder for Secure Image Steganography

Published · Scopus Indexed
Python TensorFlow Keras OpenCV Flask

Traditional steganography methods like LSB insertion are easy to attack — they have low payload capacity and are detected quickly by steganalysis tools. This paper proposes a CNN-based encoder-decoder that hides data inside images with nearly undetectable visual change, trained on 10,000 images to handle real-world distortions like compression, noise, and scaling.

The encoder extracts deep spatial features from the cover image and embeds the secret payload across them. The decoder reverses this precisely, even after the image has been manipulated. A composite loss function balances two objectives simultaneously — keeping the image visually identical to the original (MSE) while maximising data recovery accuracy (Binary Cross-Entropy).

38.2 dB PSNR — well above the 30 dB quality threshold
0.95 SSIM score — images visually indistinguishable from originals
92% Data retrieval accuracy after JPEG compression

Steganalysis detection rate kept below 30% — meaning adversaries struggle to even confirm hidden data exists. The system also maintained 88% accuracy under Gaussian noise and 85% after 50% image scaling, with a maximum payload of 1KB per image.

Thoufeeq M., Monish M.S., Lakshmi M.A.  ·  ACT 2025 — 16th International Conference on Advances in Computing, Control, and Telecommunication Technologies View on Scopus →
NLP · Machine Learning

Advanced Email Spam Detection

Python NLP Scikit-learn

Spam evolves fast. A model trained on last year's campaigns gets outpaced quickly because spammers change phrasing, structure, and tactics. This project was designed around that core challenge.

Rather than relying only on word frequencies, the pipeline extracts structural signals — header anomalies, link density, sentence length patterns, and tone markers — that stay predictive even as specific spam language changes. The classifier was tuned for both accuracy and throughput so it stays practical at email scale.

Skills

Programming

  • Python
  • R
  • SQL

Machine Learning

  • TensorFlow
  • PyTorch
  • Scikit-learn
  • Computer Vision
  • NLP

Tools

  • GitHub
  • VS Code
  • Tableau
  • Figma

Web

  • HTML
  • CSS

Achievements & Certifications

Competition wins

  • 2nd place — Agni's Ignite India Innovation Conference, BONN Nehru Hr Sec School
  • 3rd place — Codechef Technical Quiz, Sathyabama University
  • 2nd place — Cybug Online Technical Quiz, Sathyabama University

Certifications

  • Digital 101 Journey — NASSCOM Sept 2023 – Feb 2024
  • Digital 101, 30 hours — NASSCOM June – Aug 2024
  • AWS Python — Credo Systemz Jan – Apr 2024

Get in touch

I'm open to research collaborations, internships, and project discussions. Email is the best way to reach me.

monishsaravanan1201@gmail.com