Research to Real-World AI
Hi! I'm Devesh, an ML Engineer specializing in Privacy-Preserving AI for Finance.
Pursuing an MSc in Financial Technology (Intelligent Process Automation) at NTU Singapore, after a B.Tech in AI/ML (Honours, 8.89 CGPA) from University of Mumbai. 2 research papers under review at Springer. Currently building federated and graph-based systems for financial intelligence and AML.
My Tech Stack
AI enthusiast with a passion for machine learning and intelligent systems.
Languages & Data
ML / AI
Finance & Privacy
Tools & Frameworks
Projects
A selection of my recent projects and innovations in AI/ML.
Violence Detection System
Finalist at Rajasthan Police Hackathon. Real-time security system that detects weapons and violence in live camera feeds, sending alerts with location and timestamp to authorities.
BOTANIX
Finalist at Gujarat Hackathon. AI-powered plant identification system that scans plants and provides detailed information about their medicinal properties and traditional uses.
MedAI Assistant
Federated Learning system for early diagnosis of lung cancer and brain tumors from CT scans and MRI. Achieved 90%+ accuracy with secure architecture for medical deployment.
AI Image Generator
Web-based image generation tool using FLUX schnell LLM. Users can generate high-quality images from text prompts with an intuitive interface.
Smart OCR System
Invoice processing system using Qwen2vl-7b for visual document analysis. Extracts structured data into JSON format with deployed API for enterprise use.
Multi-Input Language Translator
Versatile translation tool supporting file uploads, text input, and voice input. Built with Llama model for accurate translations across multiple languages.
Experience
ML Engineer
F.T. Solutions Pvt. Ltd.
June 2025 - June 2026→Built an LLM-based invoice extraction system to automate financial document processing
→Contributed to FT-DataMind, an internal AI platform providing zero-code access to live SAP data
→Worked across the stack using React, LLM pipelines, DuckDB, and enterprise data sources
Intern
NET WORLD Technologies
Dec 2024 - March 2025→Completed 3-month intensive internship in software development
→Built web applications using modern frameworks and contributed to client projects
→Gained practical experience in project management and team collaboration
Technical Head
SORT & Literary Club
2022 - 2024→Led technical initiatives including full-stack website development for the club
→Delivered 2 professional digital magazines with cross-functional team leadership
→Improved online engagement and event participation through digital strategy
About Me
I'm an ML Engineer currently pursuing an MSc in Financial Technology (Intelligent Process Automation) at NTU Singapore, after a B.Tech (Honours) in AI & ML from University of Mumbai specializing in Artificial Intelligence in Societal Benefits. My research focuses on federated learning and privacy-preserving systems — with 2 papers under review at Springer. Beyond coding, I enjoy reading books like “The Psychology of Money” and “The Shiva Trilogy” by Amish.
I believe in using AI to solve real compliance and privacy challenges — from GDPR-compliant medical imaging to zero-knowledge verified financial intelligence. Alongside the MSc I'm working through the CFA Level I curriculum, so the systems I build rest on real finance fundamentals. My goal is to build intelligent solutions that automate manual work and eventually start my own company providing AI services in the finance sector.
Currently Working
Building innovative AI solutions and cutting-edge applications
ZK-FINet — Zero-Knowledge-Verified Federated Intelligence Network
Privacy-Preserving AIDesigned a federated learning framework integrating SCAFFOLD optimization with Zero-Knowledge Proofs to verify training correctness without exposing institutional data. Enabling privacy-preserving, regulator-auditable collaboration for financial intelligence tasks such as AML and fraud detection. System architecture aligned with real-world financial regulations and compliance constraints.
Technologies
Current Project
Implementing ZK-proof verification for distributed model training
Learning Resources
- ZK-SNARK protocols
- Financial compliance frameworks
- Federated learning security papers
MuleWatch — Real-Time Money-Mule Detection for APP Scams
AML / Financial CrimeA full-stack anti-money-laundering system that flags money-mule accounts in near real time, before scam proceeds are cashed out. Combines behavioural signals (rapid pass-through, new or dormant accounts, fan-in/fan-out) with transaction-graph features, and compares a rule-based baseline, LightGBM and an optional GNN on IBM AMLworld plus an original APP-scam simulator. Every alert carries plain-English reason codes and a FATF typology tag, and an analyst console drafts a Suspicious Transaction Report (STR). Evaluated on business metrics: PR-AUC, recall at a fixed alert budget, time-to-detect and dollars recoverable.
Technologies
Current Project
Phase 1: project skeleton, APP-scam and mule-typology simulator, and AMLworld data loader
Learning Resources
- FATF money-mule typologies
- MAS AML/CFT guidance
- IBM AMLworld (NeurIPS 2023)
- UK PSR APP-scam reimbursement rules
Global Collaboration
Building relationships across time zones and connecting with teams worldwide
Flexible Time Zones
Available across multiple time zones
Prioritize Collaboration
Team-first approach to development
Available Worldwide
Global reach and accessibility
Quick Responses
Fast communication and delivery