An IT undergrad (B.Tech) student from Kolkata, India and a member of Google's Women Techmakers since 2025 and IEEE Robotics and Automation Technical Committee on Machine Learning for Automation.
I'm an aspiring ML/DL Engineer with strong research interests in Artificial Intelligence and Machine Learning, exploring how intelligent systems can be applied to real-world challenges. Alongside my passion for AIML, I have hands-on experience in Web Development and enjoy problem-solving through Data Structures and Algorithms.
"Currently seeking opportunities in Software Engineering and Machine Learning, and additionally in Data Science, where I can build impactful, data-driven solutions."
Research & Publications
Architectural Paradigms for Sovereign AI: Mitigating API Fragility through Quantized Low-Rank Adaptation and 1.58-bit Ternary Models
A rigorous methodology for deploying Sovereign AI through Small Language Models (SLMs) to mitigate API fragility. Investigates the synergy between 4-bit NormalFloat (NF4) quantization, native 1.58-bit ternary architectures, and Low-Rank Adaptation (LoRA) for specialized domains.
- ▹Achieved a 93% reduction in inference OpEx for 7B–14B parameter models.
- ▹Demonstrated a TCO break-even at 450 million tokens vs. proprietary APIs.
Isolating Synthetic Fingerprints: A Frequency-Domain Approach to Robust Deepfake Detection using 2D FFT and Lightweight CNNs
A novel, lightweight deepfake detection engine that isolates synthetic fingerprints in the frequency domain. Employs a micro-blur preprocessing pipeline to suppress superficial high-frequency spatial noise, followed by a 2D FFT to extract underlying spectral magnitude maps.
- ▹Achieved 99.12% accuracy on FaceForensics++ and 98.45% on Celeb-DF v2.
- ▹Outperforms spatial baselines at 45 FPS real-time inference on edge devices.
Some of My Projects
- ▹Frequency-domain AI engine using 2D FFT and ResNet-18 to identify synthetic fingerprints invisible to standard spatial-only CNNs.
- ▹Achieved <1.8s end-to-end inference latency via distributed React/FastAPI/Docker stack.
- ▹Integrated 3×3 Gaussian Micro-blur filter to neutralize sensor noise.
- ▹Hybrid AI routing engine orchestrating local ChromaDB retrieval, live web-search, and direct LLM inference.
- ▹Reduced local memory overhead by 75% by offloading embeddings to the cloud.
- ▹Asynchronous RAG pipeline using FastAPI BackgroundTasks for zero-lag UI.
- ▹DDQN agent mastering Atari Space Invaders directly from raw pixel input.
- ▹CNN pipeline extracting complex spatial features from raw emulator frames.
- ▹Python CLI grouping biodiversity observations into clusters on 5km/3hr spatio-temporal thresholds.
- ▹Incremental sync engine tracking state via timestamps, reducing API latency.
- ▹Dynamic cluster merging logic for overlapping observation windows.
- ▹Full-stack React + FastAPI app using Robust PCA to extract clean text from degraded documents.
- ▹95% automated parameter tuning via custom Adaptive Regularization heuristic.
- ▹Full document restoration in under 2.5 seconds.
- ▹6-voice polyphonic VST3 synthesizer built from scratch in C++ with the JUCE framework.
- ▹4-pole Moog-style ladder filter, BBD stereo chorus, and independent ADSR envelopes.
- ▹Real-time FFT spectrum analyzer with custom UI components.
- ▹End-to-end ML model predicting customer churn — 67% recall rate and AUC 0.82.
- ▹Interactive Streamlit web app for real-time predictions deployed publicly.
- ▹Real-time procedural 3D mushroom lifecycle simulation using Three.js.
- ▹Space Colonization Algorithm with 3D Spatial Hash Grid handling 40,000 mycelial nodes.
- ▹Cinematic post-processing: Unreal Bloom, procedural geometry generation.
- ▹Computer vision tool to algorithmically detect and calculate rotational skew of images.
- ▹Tested on satellite datasets where geo-referencing introduces rotational artifacts.
- ▹Automated image registration using OpenCV to identify and map spatial keypoints across image pairs.
- ▹Vector-distance keypoint matching for mathematically accurate coordinate mappings.
- ▹Text and voice-based intelligent search system for satellite data ordering.
- ▹NLP parsing of satellite-specific terminology to interface with databases.
- ▹In-browser 3D peony flower with continuous procedural blooming/wilting lifecycle.
- ▹Custom 3D pipeline using rotation matrices and Painter's Algorithm — no WebGL.
- ▹Multi-modal post-processing: solid pixels → particle dots → real-time ASCII mapping.
- ▹Decentralized academic authenticity platform combining OCR & AI for automated data extraction with Ethereum Blockchain for immutable record-keeping.
- ▹MetaMask wallet integration for authenticated and tamper-proof certificate issuance. Top finalist out of 250+ teams in SIH 2025 internal round.
Experience
Member of Google's Women Techmakers
Google · Remote
Engage in a global community supporting women in tech through events, learning resources, and networking.
Blockchain Risk Developer
Zetheta Algorithms Private Limited · Remote
- ▹Engineered a vulnerability scanner for Solidity contracts detecting critical flaws like Reentrancy and Access Control.
- ▹Developed an algorithmic market-making engine using the Avellaneda-Stoikov model to optimize bid-ask spreads and manage inventory risk.
Open-Source Contributor
GSSoc '25 · Remote
Resolved bugs and integrated new feature modules across 4 open-source repositories via GitHub.
Web Development Intern
CodSoft · Remote
Developed and enhanced 6+ responsive web pages using HTML, CSS, and JavaScript, significantly improving UI accessibility.
Technical Arsenal
Languages
Libraries & Frameworks
Tools & Platforms
Memberships & Affiliations
Education
Techno Main Saltlake
Bachelor of Technology
Achievements & Certifications
HackerRank
Gold Level (5 Star) in Python, Silver Level (3 Star) in Java.
SIH '25 Semi-Finalist
'Digi-Pramaan' — Top finalist out of 250+ teams in the college internal round of Smart India Hackathon 2025.
Problem Solving
Solved 350+ problems on LeetCode and HackerRank, improving algorithmic thinking and coding efficiency.
Extracurricular Leadership
Active member of Samaritans social club since first year — community initiatives, collaborative events, debates, and quiz competitions.
Certifications
- ▹Data Analytics Job Simulation — Deloitte Australia
- ▹Software Engineering Job Simulation — JPMorgan Chase & Co.
- ▹Introduction to Agent Skills — Anthropic Academy
- ▹Train & Manage ML Model with Azure — Microsoft
- ▹Understand Data Science for ML — Microsoft
- ▹Python for Data Science — IBM
- ▹GenAI Powered Data Analytics Simulation — Tata
Resume
Want a detailed breakdown of my academic background, technical skills, and professional experience? View or download my complete resume below.
DOWNLOAD RESUME.PDFInitialize Connection
Always open to discussing software engineering opportunities, ML research, and innovative ideas. Let's execute a handshake.
PING ME© 2026 Anushka Das · End of File.