Skip to content

Building agentic AI for fraud & AML

Deepratna Awale

Senior Software Engineer @ Nasdaq

St. John's, NL, Canada

Building production agentic AI that helps bank BSA analysts spend their time on real fraud and money laundering, not noise.

Resume PDF Email GitHub LinkedIn Open terminal
01 / about

About

I like turning research-grade models into reliable, fast, observable systems: data pipelines, prompt engineering, evaluation, and the cloud infrastructure that lets other engineers ship agents safely.

Role
Senior Software Engineer @ Nasdaq
Based in
St. John's, NL, Canada
Certification
AWS Certified Machine Learning Engineer, Associate
Education
MASc Computer Engineering, Memorial University of Newfoundland
02 / experience

Experience

  1. Senior Software Engineer

    Nasdaq (Verafin)

    May 2026 – PresentSt. John's, NL, Canada

    Product: Verafin Agentic AI Workforce

    • Building an end to end, production ready agent creation pipeline and process: data preprocessing, prompt engineering, and infrastructure that lets other developers easily deploy agents with tools and skills.
    • Optimized an agent to increase specificity and recall while cutting its response time in half.
    • Developing agents on AWS Bedrock AgentCore that autonomously work BSA/AML cases and recommend Acknowledge or Investigate dispositions, cutting false positives so analysts spend their attention on real fraud.

  2. Generative AI Associate

    Innodata Inc.

    August 2025 – May 2026Toronto, ON

    • Evaluated and rated AI model outputs for quality, relevance, and accuracy for Meta.
    • Contributed to open-source tooling like Redlite for toxicity testing and benchmark metrics.
    • Supported dataset development through data collection and augmentation to reduce overfitting.
03 / projects

Projects

Pulled live from GitHub. Each card links to the code and, where there is one, a live demo.

open-wallpaper-engine-mac

Swift

  • macOS player for Wallpaper Engine wallpapers with native Metal renderer supporting scenes, videos, web content, particles, 3D models and SceneScript.
  • Translates Wallpaper Engine GLSL shaders to Metal automatically, caches results, and skips problematic shaders without crashing.
  • Browses and downloads wallpapers from Steam Workshop, imports local files, and supports audio-reactive visuals and MCP server control.
CodeLive

AutoExpress

JavaScript★ 27

  • Generates 28 different facial expressions from input images using Stable Diffusion inpainting with ADetailer extension
  • Processes both anime and realistic character images to create emotion grids automatically
  • Provides web UI to configure prompts and generation parameters, integrates with A1111 Stable Diffusion WebUI API
Code

3t-chatbot

TypeScript

  • LLM-powered chatbot answering questions about Deepratna Awale's work and portfolio using Next.js and Vercel AI SDK.
  • Integrates real-time GitHub repository data with automated daily profile refresh for recruitment-focused conversations.
  • Supports multiple LLM providers including xAI Grok, OpenAI, and others with persistent chat history via Neon Postgres.
CodeLive

sd-parsers

TypeScript

  • Extracts prompt information and generation parameters from Stable Diffusion images created by Automatic1111, Fooocus, ComfyUI, InvokeAI, and NovelAI.
  • TypeScript package that parses embedded metadata from AI-generated images and returns structured prompt data with helper functions.
  • Supports multiple extraction effort levels and configurable parser modules for flexible metadata retrieval from image files or buffers.
CodeLive

TAES2

Python

  • Evaluates theoretical answers using LLMs and RAG technology with support for GPT, Claude, Gemini, and Ollama models.
  • Processes question banks from PDF/DOCX files and batch evaluates up to 100 answer sheets with detailed feedback and analytics.
  • Provides three interface modes (Main, Simple, Minimal) with PostgreSQL database, student management, and comprehensive evaluation history.
Code

Polar-Image-Inspector

Jupyter Notebook

  • Import and view metadata from Polar Radar Image .pol files with support for 8-bit and 16-bit data types.
  • Convert polar radar byte data into cartesian images with configurable rendering options.
  • GUI application with dark mode to inspect radar images and save output with embedded metadata.
Code

All repositories on GitHub

04 / skills

Skills

Agentic AI

  • Bedrock AgentCore
  • LangChain
  • prompt engineering
  • evals
  • RAG

ML / DL

  • PyTorch
  • TensorFlow
  • scikit-learn
  • OpenCV
  • diffusion models

Cloud / Infra

  • AWS
  • Terraform
  • Docker
  • CI/CD
  • Lightsail
  • SageMaker

Data

  • PySpark
  • pandas
  • ETL pipelines
  • Neo4j
  • SQL

Languages

  • Python
  • TypeScript
  • Java
  • C++
  • SQL
05 / publications

Publications

IRJCS, 2021

Semantic Analysis of Long Answers

Evaluates long-form exam answers by encoding sentences with a Deep Averaging Network and comparing them to an answer key.

academia.edu ResearchGate

IJSRP, 2022

Theoretical Answer Evaluation System [T.A.E.S]

NLP system for automated scoring of theory answers with plagiarism detection and grammar penalties.

academia.edu
06 / education

Education & certifications

MASc, Computer Engineering

Memorial University of Newfoundland · 2024

Thesis work: hybrid CNN + MLP to estimate ocean wave height from Wamos II radar images.

BEng, Information Technology

RGCER, Nagpur · 2021

  • AWS Certified Machine Learning Engineer, Associate2025
  • IBM Data Science Professional Specialization2019
  • IIT Madras Programming and DSA Using Python2019
07 / now

Now

Updated October 2026

  • Building agents on AWS Bedrock AgentCore at Nasdaq (Verafin) that work BSA/AML cases and cut false positives.
  • Building the pipeline that lets other engineers ship agents with tools and skills safely.
  • Running this portfolio in the open: Terraform, Lightsail and Bedrock. Source: GitHub.
08 / guestbook

Guestbook

Leave a note: the guestbook needs JavaScript, or sign it from the terminal with guestbook sign.

09 / contact

Contact

Let's build something reliable.

Email is the fastest way to reach me.

awale.deep@gmail.com Resume