Debjyoti Ghosh

AI Security Researcher & ML Engineer

Undergraduate researcher and AI Security engineer specializing in agentic AI trust & governance, multimodal deepfake detection, voice biometrics anti-spoofing, LLM security, and trustworthy AI systems. Published at IEEE INDICON 2025. Experienced in context-aware NLP, media forensics, and scalable backend infrastructure.

Download CV (PDF)Research Papers

Scroll to explore

Technical Expertise

Skills & Technologies

Tools, frameworks, and programming languages I use to build research models and production backend infrastructure.

Python
Python
C
C
SQL
SQL
Bash
Bash
TypeScript
TypeScript
JavaScript
JavaScript
PyTorch
PyTorch
TensorFlow
TensorFlow
Hugging Face Transformers
Hugging Face Transformers
Scikit-learn
Scikit-learn
Bayesian Neural Networks (BNN)
Bayesian Neural Networks (BNN)
SBERT
SBERT
DistilBERT
DistilBERT
FastAPI
FastAPI
Flask
Flask
PostgreSQL
PostgreSQL
RESTful APIs
RESTful APIs
asyncio
asyncio
ETL Pipelines
ETL Pipelines
Microservices
Microservices
Pandas
Pandas
NumPy
NumPy
OpenCV
OpenCV
Librosa
Librosa
MFCCs
MFCCs
Mel-spectrograms
Mel-spectrograms
Vector Embeddings
Vector Embeddings
Git
Git
Docker
Docker
AWS
AWS
VS Code
VS Code
Research & Publications

My Research Work

Cutting-edge research in AI/ML with focus on deepfake detection, forensics, and biometric applications

Featured Research22nd IEEE INDICON 2025

Multimodal Deepfake Detection via Audio-Visual Fusion

95% (0.97 AUC) Accuracy

Architected a scalable multimodal deepfake detection pipeline utilizing a Transformer-based attention mechanism to fuse spatial, temporal, and acoustic embeddings, achieving 95% accuracy and 0.97 AUC.

Key Highlights

  • Integrated ResNet-50 visual encoder, BiLSTM temporal encoder, and 1D-CNN audio encoder processing Mel-spectrograms.
  • Implemented custom synchronization loss function to enforce phoneme-viseme alignment verification.
  • Evaluated on FakeAVCeleb benchmark dataset demonstrating superior cross-modal robustness.

Technologies

PyTorchResNet-50BiLSTM1D-CNNMel-spectrogramsTransformers
Read Paper (PDF)IEEE DOIRepository
Accepted Research Paper

Context-Aware Hate Speech Detection via Post-Comment Joint Modeling

95% (0.9911 AUC) Accuracy

Created a context-aware moderation framework replacing isolated text analysis with post-comment joint modeling, achieving 95% accuracy and ROC-AUC of 0.9911.

Key Highlights

  • Structured four-class reaction taxonomy mapping discourse to explicit moderation actions.
  • Fused parallel features including SBERT semantic similarity, emotion distributions, and DistilBERT probabilities.
  • Validated framework robustness on curated social media interaction dataset.

Technologies

PythonSBERTDistilBERTNLPPyTorchScikit-learn
Read Paper (PDF)Repository
AIFB 2025 / University Project

Edge-Based Voice Biometric Verification & Cloud Anti-Spoofing

97.5% Accuracy

Constructed a speaker recognition system combining edge-based audio acquisition with a centralized cloud backend, achieving 97.5% authentication accuracy in controlled environments.

Key Highlights

  • Formulated embedding pipeline utilizing MFCCs and spectral features via Gaussian Naive Bayes classifier.
  • Engineered FastAPI backend and web interface facilitating secure user enrollment and parallelized inference.
  • Evaluated against environmental noise and spoofing vulnerabilities.

Technologies

PythonFastAPILibrosaMFCCsOpenCVScikit-learn
Read Paper (PDF)Repository
Open Source & Tools

Developer Extensions & Tools

Open-source tools, IDE extensions, and developer utilities published for the global community.

VS Code Extensionv1.0.350 Downloads

AntiGravity (Dev Timekeeper) -- VS Code Extension

  • Created and published a developer productivity extension for VS Code with over 1,300 downloads on Open VSX Registry.
Open VSX Registry
Experience

Professional Journey

Backend Engineer (Python Developer)

Nov 2025 -- Feb 2026

ThreatPurge Private Limited

Engineered scalable backend services and asynchronous ETL pipelines for phishing detection and brand abuse monitoring.

Architected FastAPI & PostgreSQL microservices processing over 50,000 daily security events.
Accelerated ETL threat intelligence processing by 40% with Python asyncio.
Optimized query execution and indexing strategies, reducing DB latency by 35%.
View Certificate (PDF)

Full-Stack AI Engineer (Freelance)

Mar 2026 -- Present

Elaina Legal AI

Architected dual-facing Agentic AI legal assistant and Bayesian Neural Network (BNN) for legal outcome prediction.

Formulated AI tool-calling framework for natural language scheduling and invoicing.
Built automated OCR evidence pipelines to process client uploads in real-time.
Constructed embedding-based semantic retrieval systems for long-term memory.