SENIOR AI/ML ENGINEER
|
ABOUT
Here is a little background
Hey 👋 I'm a Senior AI/ML Engineer currently based in Mumbai, India, with 4+ years of experience building production-grade machine learning systems.
I specialize in translating complex data into actionable product insights — collaborating closely with Product and Customer-facing teams to drive real business value. My work spans AML fraud detection, transaction monitoring, and LLM-powered automation pipelines.
I'm passionate about the full model lifecycle — from high-fidelity feature engineering and graph-based fraud pattern detection to deploying scalable XGBoost, GNN, and RAG systems in production. I thrive in high-growth, foundational team environments where the work directly shapes the product.
My foundation is a B.Sc (Research) in Physics from Shiv Nadar University — and it shows in how I work. Physics trained me to strip problems down to first principles, reason under uncertainty, and build mental models before touching code. That same instinct is what drives how I approach fraud detection, model design, and system architecture today.
EXPERIENCE
Senior AI/ML Engineer
Facctum SolutionsOCT 2025 – PRESENT · PUNE, HYBRID
- Spearheaded the design and R&D lifecycle of a high-fidelity Automated KYC System, integrating PaddleOCR with a custom LLM orchestration pipeline — achieving 97.4% extraction accuracy across diverse document typologies.
- Led development of an LLM-powered automation layer for transaction monitoring, enabling seamless translation of regulatory requirements into executable monitoring rules, significantly reducing manual rule-authoring latency.
- Architecting an end-to-end AML fraud detection engine using a hybrid ensemble of Graph Neural Networks (GNNs) and XGBoost to identify "layering" and "structuring" money laundering patterns.
- Owning the full model development lifecycle including feature engineering of transactional graph metrics and hyper-parameter tuning to optimize the precision-recall trade-off for high-risk customer risk rating.
Data Scientist / ML Engineer
Mechademy Inc.DEC 2020 – OCT 2025 · REMOTE
- Led development of AutoEDA and AutoML libraries using a unified distributed system with Dask pipelines and a Ray cluster, reducing development time by 70% while ensuring 100% consistency in feature engineering.
- Built a Generative AI solution for automated configuration creation, processing unstructured PDFs through a specialized DAG architecture — reducing manual setup effort by 50%.
- Monitored and maintained production ML pipeline health using Dagster, ensuring high data quality and adherence to team SLAs.
- Owned end-to-end anomaly detection using XGBoost and statistical modeling for real-time sensor data, improving failure detection accuracy by 30%.
- Centralized multiple terabytes of historical data on AWS S3 / Azure Blob Storage and created a Delta Lake to streamline data accessibility across projects.
- Enhanced ML alerting system with SHAP/LIME model explainability, providing clear reasoning for 100% of alerts and improving decision-making for non-technical users.
SKILLS
Hover over a skill to see proficiency
PROJECTS
Project 1: Trend-to-Content Automation Engine
↗ Live DemoA Python/Streamlit app that ingests real-time trending topics from Google Trends and Reddit, enriches them with live news via web scraping, and uses Gemini/OpenAI LLMs to generate social media posts, video scripts, and AI image prompts — with one-click publishing to Twitter/X and Instagram via their respective APIs.
Deployed on AWS EC2 with secrets managed via AWS Secrets Manager (5-min TTL rotation), user authentication through AWS Cognito (sign-up, login, MFA-style confirmation flow), and a Cloudflare Workers endpoint serving a VLM-powered image generation API backed by Cloudflare's AI inference infrastructure.
Designed a modular, type-safe pipeline with Pydantic-validated data models at every stage, a dual LLM provider abstraction (Gemini/OpenAI), and a comprehensive test suite using pytest and Hypothesis for property-based testing — ensuring reliable LLM output parsing and schema conformance across the ingestion, analysis, and generation layers.
Project 2: MediVault — Serverless AI Healthcare Platform
↗ Live DemoA completely serverless platform that solves "healthcare documentation chaos" by transforming fragmented medical reports and prescriptions into an organized, instantly searchable digital repository via natural language queries. Architected an event-driven RAG pipeline with a Mistral 3 8B vision fallback, achieving 100% success rates for complex PDFs. Reduced infrastructure costs by approximately 90% while maintaining 95%+ retrieval relevance.
Project 3: Financial Advisor Agent
↗ Live DemoAn AI agent capable of providing current market sentiment for specific stocks and offering market suggestions. Integrated a CI/CD pipeline using GitHub Actions to automate testing and deployment to Hugging Face Spaces. Backed by a PostgreSQL database on Neon storing historical and real-time stock data extracted via the Zerodha API.