Machine Learning & AI Systems Architecture

Bridging the Gap Between
Scale and Reliability.

Building robust, observable Data and AI systems. Currently architecting deterministic financial systems at AWS FinTech and engineering scalable machine learning solutions.

Industry Engineering → About Me
Daksh Patel
Daksh Patel
Data Engineer · AWS FinTech
Core Expertise Data Engineering, AI Systems Architecture, Retrieval Augmented Generation
01 — About

The Narrative

My work sits at the intersection of high-stakes industrial engineering and practical AI implementation. At AWS, I lead Scenario Modeling for global cost allocation, where the margin for error is zero.

This experience, combined with my time as a Teaching Assistant for Machine Learning at USC, has driven me to focus on how we can build AI systems that are reliable, performant, and grounded in data.

My goal is to develop AI systems that are observable, verifiable, and scalable — systems that don't just predict, but provide deterministic engineering guarantees.

The Engineering Philosophy

Why can we model millions of AWS billing scenarios with deterministic precision, yet still struggle to build AI systems that genuinely align with strict engineering constraints? I aim to bridge this gap.

FEAT — 01
Data & Systems Architecture
Designing high-fidelity frameworks to process complex data at scale, with verifiable logic and deterministic guarantees.
FEAT — 02
Machine Learning Optimization
Stripping away algorithmic bloat to find the mathematical signal — what the model must retain vs. what it can safely discard for performance.
FEAT — 03
Observable AI Systems
Designing AI that operates transparently, not just as black boxes. Building observable, verifiable ML pipelines.
FEAT — 04
Applied Computer Vision
Applied deep learning for medical image classification at USC — where diagnostic precision and system reliability are life-critical.
02 — Experience

Industry Work

Jun 2025 – Present
Amazon Web Services
Seattle, WA · FinTech
Data Engineer — FinTech, Cost Allocation
Designing high-fidelity frameworks to simulate complex financial outcomes at cloud scale, focusing on system robustness and verifiable logic in mission-critical billing infrastructure.
  • Built an automated scenario modeling engine using Python & SQL for high-fidelity "what-if" simulations, reducing latency by 85% and enabling proactive impact analysis.
  • Architected migration of 20TB+ financial workloads to ATLAS with 100% integrity; automated validation frameworks resolved 99% of data discrepancies for security auditing.
  • Refactored PySpark ETL pipelines processing billions of records daily, optimizing compute efficiency by 30% with zero production downtime.
  • Modernized legacy logic into scalable SQL/Spark architectures, improving end-to-end data delivery speeds by 25%.
Oct 2024 – Jun 2025
Northern Lights Post Inc.
Los Angeles, CA
Machine Learning / Data Engineer
Developed multimodal content ranking systems bridging production-scale ML with research-grade retrieval precision at 9M+ image, 10M+ video scale.
  • Developed viral content ranking algorithm using XClip (video) and ResNet-50 (image) embeddings for popularity prediction.
  • Integrated FAISS-based similarity search for efficient embedding-based retrieval, reducing computational overhead significantly.
  • Automated metadata augmentation pipelines improving recommendation model precision and reducing manual intervention.
Jan 2024 – May 2025
University of Southern California
Los Angeles, CA
Graduate Teaching Assistant — DSCI 552
Course Producer under Prof. Mohammad Reza Rajati. Mentored 750+ students across three semesters in ML fundamentals and investigated Deep Learning applications for medical imaging.
  • Earned TA role by ranking Top 5 in the course cohort.
  • Provided personalized guidance to 750+ students across three semesters of Machine Learning for Data Science.
Nov 2023 – May 2025
Keck School of Medicine, USC
Los Angeles, CA · Radiomics Lab
Research Assistant — Oncology Imaging
Applied deep learning to medical image classification for cancer detection in a life-critical diagnostic setting — where precision is non-negotiable.
  • Optimized diagnosis pipelines with PyTorch, integrating Foundation Models (SAM, SAM2, MEDSAM) for segmentation and comparing with nnU-Net benchmarks.
  • Applied quantization to decrease model parameters and enhance computational efficiency by 15%.
  • Developed novel data preprocessing to mitigate class imbalance, boosting weighted accuracy by 10%.
May – Sep 2024
Kintsugi Global, Inc.
Los Angeles, CA
AI/ML Engineer — NLP & Voice Systems
  • Custom-trained LLMs using PyTorch for an anime-character chatbot; enhanced user engagement by 25%.
  • Engineered a TTS system using NVIDIA Tacotron 2; applied quantization and distillation to reduce latency for real-time voice generation.
  • Integrated AWS services (EC2, Lambda, S3) for scalable, low-latency inference deployment.
Jan – Sep 2023
Siksti Technologies (Slikk)
Bengaluru, India
Machine Learning Engineer
  • Leveraged ML to identify sales trends, contributing to a 10% revenue increase through data-driven decisions.
  • Reduced catalog processing time by 80% via automated backend scripts using Django and PostgreSQL.
  • Boosted API response time by 66% through caching, query optimization, and load balancing.
03 — Portfolio

Research & Engineering

Systems Engineering · AWS
Scenario Modeling Engine — AWS FinTech
Automated high-fidelity "what-if" simulation framework for cloud-scale financial cost allocation. Deterministic billing simulation across millions of scenarios with zero tolerance for error. 85% latency reduction; 20TB+ migration with 100% integrity.
PythonSQLPySparkATLAS
Proprietary System
Medical AI · Published
Deep Learning for Osseous Metastatic Cancer Detection
Developed and evaluated DL models to automate lesion detection and segmentation in CT scans at Keck School of Medicine. Integrated Foundation Models (SAM, SAM2, MEDSAM) with nnU-Net benchmarking for oncology imaging.
PyTorchSAM2nnU-NetMedical Imaging
View Publication →
Multimodal · Retrieval
FAISS-Accelerated Content Ranking System
Viral content ranking engine using XClip (video) + ResNet-50 (image) embeddings with FAISS-based similarity search. Optimized embedding-based indexing for sub-linear retrieval at 9M+ image, 10M+ video scale.
FAISSXClipResNet-50Embeddings
Proprietary
NLP · Architecture
Transformer from Scratch
Decoder-only Transformer built entirely from scratch with PyTorch and PyTorch Lightning. Implements Multi-Head Self-Attention, Feed-Forward Networks, Positional Encoding, and Layer Normalization — no library abstractions.
PyTorchTransformersAttention
View on GitHub →
NLP · Translation
TransLingo: Neural Machine Translation
Seq2Seq translation with attention mechanisms for German→English. Focuses on representation alignment between source and target language spaces through encoder-decoder architecture.
PyTorchSeq2SeqAttention
View on GitHub →
Remote Sensing · Published
Predicting Agricultural Yield via Remote Sensing
ML model for Punjab crop yield prediction using 17 years of multi-source geospatial data (weather, soil, NDVI satellite data). Statistical methods, ML algorithms, and real-world validation. Published ICoISS 2023.
Remote SensingNDVIStatistical ML
View Publication →
04 — Research Output

Selected Publications

01
European Journal of Radiology AI · 2025
Deep learning-based detection and segmentation of osseous metastatic prostate cancer lesions on computed tomography
Automated lesion detection and segmentation in CT scans using deep learning, demonstrating AI-driven automation improving metastatic lesion detection accuracy and clinical decision-making in oncology.
doi:10.1016/j.ejrai.2025.100005 ↗
02
ICoISS · June 2023
Predicting Agricultural Yield by Integrating Remote Sensing Data and Machine Learning Technology
Multi-source geospatial yield prediction for Punjab farmlands using 17 years of weather, soil, and NDVI satellite data with statistical and real-world validation.
doi:10.1007/978-981-99-1726-6_6 ↗
03
Procedia Computer Science · January 2022
Blockchain-based Food Supply Chain — A Double Blockchain Framework
Double-blockchain framework to enhance transparency, traceability, and trust in the agricultural supply chain with immutable audit trails and provenance verification.
doi:10.1016/j.procs.2022.12.034 ↗
05 — Toolkit

The Engineer's Toolkit

Core Engineering
Data Architecture
System Optimization
Statistical Modeling
Scenario Modeling & Simulation
Distributed Systems
CI/CD Pipelines
Test-Driven Development
AI & Systems
PyTorch
Transformers / HuggingFace
vLLM
TensorRT-LLM
Quantization & Distillation
FAISS
TensorFlow · Keras · Scikit-learn
OpenCV · SpaCy · LangChain · MLFlow
Data & Cloud
PySpark / Apache Spark
AWS (EC2, S3, SageMaker, Lambda, RDS)
SQL · PostgreSQL · NoSQL
Hadoop · Kafka · Apache Airflow
Docker · CI/CD
Python · R · Java · C++ · MATLAB
Tableau · Power BI · Excel
Open to Opportunities · ML Engineering

Currently exploring roles in
Machine Learning and Data Engineering.

Let's discuss the intersection of deterministic systems, information theory, and human-aligned machine learning. I document my thinking at Medium.

Get in Touch → Read the Blog ↗