SS
Booting Portfolio
Open to new opportunities · NYC

Most ML models
die in a notebook.
Mine ship to production.

Hey, I'm Shashank Reddy Sama — a Machine Learning Engineer at JPMorgan Chase, building fraud detection, LLM systems, and time-series forecasting at scale. I live in the messy middle between research and production.

Experience4+ years
Based inBrooklyn, NY
CurrentlyJPMorgan Chase
SS

Shashank Reddy Sama

ML Engineer · 900 followers

98%
Fraud detection accuracy
38%
Reduction in fraud volume
35%
Forecast accuracy lift
5TB+
Text data processed
PyTorch LLMs AWS SageMaker FastAPI MLOps
About

The space between research
and production.

"

The part where a beautiful model meets a real dataset, a latency budget, and a regulatory requirement — and has to still work. That's the problem I want to solve for the rest of my career.

I'm a Machine Learning Engineer at JPMorgan Chase, where I build the systems that help one of the world's largest banks detect fraud, forecast financial outcomes, and make smarter decisions at scale.

Before JPMC, I was at Accenture building LLM-powered chatbots and NLP pipelines. Before that, I earned my MS in Mechatronics & Robotics at NYU — where I got obsessed with teaching machines to see and make decisions in the real world.

I'm always down to talk about deploying ML responsibly in high-stakes environments, the jump from robotics research to production ML, and why MLOps is the most underrated skill in the industry.

4+
Years of ML engineering across finance & enterprise
13
Shipped projects in CV, NLP & robotics
3
Companies — JPMC, NYU Research, Accenture
$9k
Annual NYU merit scholarship recipient
Skills

A toolkit built for shipping.

Production ML isn't one skill — it's the unglamorous combination of modeling, infrastructure, data engineering, and the patience to debug a 4 a.m. inference timeout.

Modeling & Frameworks
PyTorch95%
TensorFlow / Keras90%
Hugging Face Transformers92%
Scikit-learn / XGBoost88%
Deep Learning & LLMs
CNN / LSTM / Transformers94%
BERT & Llama fine-tuning90%
RAG & LangChain pipelines85%
Computer Vision (YOLO, OpenCV)88%
Cloud & MLOps
AWS (SageMaker, Lambda, EC2)93%
FastAPI / WebSockets90%
GCP (Vertex AI)82%
CI/CD & Docker86%
Robotics & Systems
ROS / ROS2 & Gazebo85%
Visual SLAM & Sensor Fusion80%
Path Planning (A*)82%

The full stack

Languages, frameworks, infra, and data tools I reach for daily.

Languages
PythonC++Java C#JavaScriptMATLAB React
ML & Data
NumPyPandasPySpark NLTKOpenCVMatplotlib SeabornA/B Testing
Cloud & Infra
EC2S3SageMaker LambdaCloudWatchAPI Gateway Vertex AIDocker
Databases
PostgreSQLMySQLSQL Server MongoDBRedisNeo4j
Visualization
TableauPower BI JupyterGoogle Colab
Experience

Four years of shipping ML to prod.

From enterprise chatbots to fraud detection running on millions of daily transactions — every role taught me something about the gap between a model that works and one that scales.

Machine Learning Engineer
JPMorgan Chase & Co.
New York, NY · Full-time
Mar 2024 — Present
  • Shipped an LSTM-based fraud detection model to production with 98% accuracy — now part of JPMC's compliance and transaction-monitoring stack flagging suspicious activity across millions of daily transactions.
  • Cut fraudulent transaction volume by 38% by architecting CNN-based anomaly detection systems that catch patterns legacy rule-based systems missed.
  • Reduced model deployment time by 30% by engineering FastAPI inference pipelines on AWS (EC2, Lambda, SageMaker) and GCP — turning 2-week deployments into 4-day ones.
  • Improved financial forecasting accuracy by 35% through ensemble modeling, driving sharper capital allocation decisions.
  • Boosted NLP model performance by 10% via BERT fine-tuning and GAN-based data augmentation for text analytics.
  • Doubled inference speed by building backend systems leveraging FastAPI, WebSockets, and multiprocessing for real-time bidirectional transaction monitoring.
PyTorchLSTMBERT FastAPISageMakerMLOps GANs
Research Assistant — Conversational AI
New York University
New York, NY
Jan 2023 — May 2024
  • Built a Conversational AI system using fine-tuned Llama and BERT (PyTorch + Hugging Face) that improved context-aware response accuracy by 22% over the baseline.
  • Processed 5TB+ of unstructured text using PySpark + advanced NLP (NER, tokenization, sentiment) — pushing entity extraction precision up 18% and cutting model training time by 40% via better feature pipelines.
  • Deployed optimized NLP models on AWS for real-time inference, dropping end-to-end latency by 35% and proving the architecture could scale beyond the lab.
  • Conducted Proof of Concepts to validate scalable ML architectures, designing robust pipelines for data preprocessing and feature engineering.
LlamaBERTHugging Face PySparkNLPAWS
Machine Learning Engineer
Accenture
Hyderabad, India · Full-time
Feb 2021 — Jul 2022
  • Built and deployed an LLM-powered enterprise chatbot using RAG architectures + Confluence APIs that lifted customer service efficiency by 25%.
  • Shrunk model size by 38% without sacrificing accuracy through quantization and pruning, making NLP and CV models deployable on tighter compute budgets.
  • Cut production incident resolution time by 30% by deploying scalable models on AWS with CloudWatch-based monitoring and alerting.
  • Created automated OCR + RAG pipelines using OpenCV, improving text extraction accuracy to 98% and reducing retrieval latency across enterprise systems.
  • Fine-tuned BERT, RNN, and CNN models for intent classification, sentiment analysis, and anomaly detection — regularly beating internal benchmarks.
RAGBERTOpenCV QuantizationCloudWatchPandas
Selected Projects

Things I've actually built.

A mix of NYU capstones, undergrad research, and personal explorations across computer vision, NLP, and robotics. The unifying thread: making machines perceive and decide in messy, real environments.

🦾

Automated Garbage Sorting

Capstone at NYU. Real-time waste classification using YOLO + OpenCV, driving 6 xArm robotic arms positioned along a conveyor belt to sort recyclables vs. non-recyclables.

YOLOOpenCVxArmRobotics
NYU · Capstone
🧭

Vision-Based Maze Navigation

Autonomous robot that navigates an unknown maze using Visual SLAM for mapping, sensor fusion for perception, and A* path planning to find the shortest route on subsequent runs.

Visual SLAMA* PlanningSensor FusionRviz
NYU · Robotics
📦

Warehouse Automation w/ Networked Robots

Simulation of a swarm of warehouse robots in Gazebo + ROS2. Designed coordination algorithms for task distribution, inventory management, and order fulfillment under realistic spatial constraints.

ROS2GazeboSwarm Robotics
NYU · Simulation
🧠

Brain Tumor Detection & Segmentation

CNN-based pipeline for tumor detection on MRI scans. Benchmarked ResNet, VGG, and custom architectures — VGG won. Added a segmentation step to highlight tumor regions for diagnosis support.

CNNResNetVGGMedical Imaging
IIIT · 2021
😷

Live Face Mask Detection

Real-time face mask detection on live video feeds using CNN + YOLO, augmented with ANN for decision-making. Built during the pandemic for public-health screening checkpoints.

CNNYOLOReal-time CV
Computer Vision
🛞

Self-Balancing Robot

A Segway-style two-wheeled robot built on the inverted pendulum principle. PID-controlled balance, machine vision for obstacle avoidance, and a custom web app for remote control over a Raspberry Pi.

PID ControlRaspberry PiWeb App
IIIT · Robotics
📍

Indoor Positioning System

WiFi-trilateration-based positioning for indoor environments where GPS fails. Real-time signal-strength processing for navigation in malls, hospitals, and office buildings.

WiFi TrilaterationSignal Processing
IIIT · Research
🤖

SCARA Manipulator Trajectory Tracking

High-precision trajectory tracking for a SCARA robotic arm. Modeled kinematics + dynamics in MATLAB, simulated in Simulink, and tuned controllers for assembly-line accuracy.

MATLABSimulinkRobot Kinematics
NYU · Control Systems
🚗

Automated Guided Vehicle

An autonomous wheeled robot for warehouse logistics. Follows RFID-tagged paths, detects obstacles, and stops safely. Built on Arduino + IoT for distributed coordination.

ArduinoRFIDIoT
IIIT · Automation
👁️

Iris Segmentation for Biometrics

Efficient iris boundary detection in MATLAB — isolating iris from surrounding ocular structures for biometric ID systems. Optimized for accuracy under tight latency budgets.

MATLABImage ProcessingBiometrics
IIIT · Vision
📉

Quantization of Dense Neural Networks

Research into reducing the memory footprint and latency of dense neural networks via quantization, while preserving accuracy — an essential trick for edge deployment.

QuantizationDeep LearningOptimization
Research
🛡️

Web-Controlled Security Robot

Remote-controlled surveillance robot with live video streaming via a custom web app. Raspberry Pi core, machine vision for movement detection, and IoT for over-the-internet control.

Raspberry PiMachine VisionWeb App
IIIT · IoT
Education

Where I learned to think.

🗽
New York University
MS — Mechatronics & Robotics
Sept 2022 — Apr 2024
$9k merit scholarship

Capstone: Automated garbage sorting (YOLO + 6 xArm robotic arms). Coursework spanned ROS2, Visual SLAM, deep learning, and computer vision.

🏛️
IIIT Sricity
BTech — Electrical Engineering
Undergraduate
Robotics & ML focus

Foundations in Python, TensorFlow, signal processing, and embedded systems. Built brain tumor detection, AGV, and IPS systems here.

📚
FIITJEE
Intermediate — MPC
Pre-University
Grade: 97%

Mathematics, physics, and chemistry. The grind that taught me that hard problems usually just need a clean first principles attack.

🏫
Johnson Grammar School
10th Class — ICSE & IB
Secondary
Grade: 92%

Where the curiosity for how things work first kicked in. School competitions, science fairs, and a lot of disassembled household electronics.

Languages & Communication

Comfortable working across multilingual teams.

EnglishProfessional · Fluent
HindiNative
TeluguNative
SS
Contact

Let's build something that ships.

If you're hiring, collaborating on a side project, or just want to swap notes on fraud detection, robotics, or computer vision — my DMs are open. Always down to talk shop.