The Intelligence Matrix
A multi-dimensional constellation of my engineering expertise.
Hover nodes for insights & click for deep-dive associations.
Mittul Daswani
AI Engineering
Data Architecture
Systems Design
Python
Custom LLMs
RAG & LangChain
PyTorch & DL
Computer Vision
NLP & Spacy
LangGraph
CrewAI
Time Series
TensorFlow
SQL & NoSQL
Apache Spark / PySpark
Turbovec
Data Intel
R & MATLAB
Databricks
Redis
Apache Lucene
XGBoost
C/C++ & CUDA
Cloud & DevOps
WebSocket
React / Next.js
Docker
Microsoft Azure
Terraform
Tailwind CSS
TypeScript
Playwright
Professional Journey
A timeline of my experience in driving innovation, developing AI systems, and extracting actionable insights.
TWG AI twg.ai
AI Engineer
June 2025 - Present
TWG AI twg.ai
AI Engineer
June 2025 - Present
- Engineered a Patent Pending, 98% accurate, production-grade Agentic RAG Pipeline for the financial sector in PySpark within Palantir Foundry Platform in 4 weeks - replacing a legacy 32-week development cycle pipeline - to serve as the primary processing engine for all projects across the company, all while strictly complying with SOC 2 compliance.
- Saved the company $500,000/month by replacing inefficient legacy infrastructure with a state-of-the-art Agentic RAG Pipeline operating at a fraction of the cost ($0.11 per page) while simultaneously boosting extraction accuracy by 15%.
- Spearheaded the multi-agent backend for the Equity Research Solution, a platform showcased to the President and CIO of Bank of America, the CEO and CTO of Guggenheim Securities, and other CIOs/CTOs of financial companies, receiving stellar feedback for its cutting-edge analysis and investment recommendations.
- Architected and implemented AI workflows for deal synthesis and surveillance monitoring, enterprise-wide retrieval pipelines for Asset Backed Securities across 5+ sectors, solving business problems for clients including Guggenheim Investments and Department of War - cutting down deal decision times by 92%.
- Built Media Asset Management Platform with custom YOLO scene detection algorithms and image classification models for detecting key scenes from broadcast sports videos, along with dubbing, topic modeling, and a discovery/recommendation engine for optimizing marketing workflows - delivered to Teton Ridge American Rodeo Team, LA Dodgers, LA Sparks, and Pittsburgh Steelers, saving media output times by 74%.
- Developed a production-grade Equity Research intelligence engine that ingests and analyzes SEC filings (13F, 10K, 10Q) and earnings calls; built a 16+ million row knowledge base for a total procurement cost of only $274.
- Designed an enterprise-wide server-side Ontology Querying Tool capable of executing complex queries against 16+ million records at low latency, optimizing retrieval times by 31%, replacing Foundry AIP tool querying.
- Optimized the RFP Project, architecting a RAG-ready data pipeline utilizing the eVestment API and building specialized agents for automated RFP screening and intelligent user querying - eliminating tech debt and optimizing speed by 15%.
Keck School of Medicine USC (Dr. Frank OCD Lab) keck.usc.edu
Research Assistant
Oct 2024 - May 2025
Keck School of Medicine USC (Dr. Frank OCD Lab) keck.usc.edu
Research Assistant
Oct 2024 - May 2025
- Participated in 2 research papers on proving DSWPD (Delayed Sleep Wake Phase Disorder) and Delayed Sleep Onset in OCD patients using Fitbit data, time series analysis, and anomaly detection - all while complying with strict HIPAA compliance.
- Analyzed physiological and behavioral data from OCD (Obsessive Compulsive Disorder) patients, utilizing Fitbit-derived health metrics and daily surveys from both OCD-diagnosed and non-OCD participants.
- Developed a scalable data integration pipeline using parallel processing and distributed data handling to merge Fitbit physiological data (heart rate variability, sleep stages, activity levels) with structured survey responses, reducing data processing time by 45% and enabling real-time analysis.
- Implemented Isolation Forests and z-score analysis to detect anomalies in health metrics, identifying unique treatment response patterns and contributing to a 30% increase in behavioral trigger identification accuracy.
- Employed time series and survival analysis models to assess physiological and behavioral responses to OCD treatment, enhancing early detection of OCD episodes by 25% and supporting proactive intervention strategies.
- Engineered advanced signal processing workflows and statistical transformations on multivariate wearable data, extracting critical physiological indicators; optimized feature space through PCA & t-SNE, unveiling deeper insights into behavioral health patterns.
- Designed real-time, interactive visualizations to dynamically monitor and compare health trends across cohorts, facilitating the research team's evaluation of treatment effectiveness and supporting data-driven decision making.
Hyphenova hyphenova.com
AI / ML Developer & Program Manager
May 2024 - Aug 2024
Hyphenova hyphenova.com
AI / ML Developer & Program Manager
May 2024 - Aug 2024
- Developed a Multimodal AI Chatbot using a fine-tuned LLM, computer vision agent, and song analysis, improving user engagement by 58% with content benchmarking against top influencers.
- Built a Profanity Detection Model with n-grams, Perspective API, and KeyBERT with 95% precision.
- Implemented a Music Recommendation System using sentiment and structure analysis of songs, increasing user satisfaction by 25%.
- Steered AI/ML strategy alongside the CEO as a core management team member, aligning projects with business goals and driving innovation.
GlobalShala globalshala.com
Data Visualization Domain Expert
May 2022 - July 2022
Project Head Internship
April 2022 - May 2022
Data Analyst
Feb 2022 - March 2022
GlobalShala globalshala.com
Data Visualization Domain Expert
May 2022 - July 2022
Project Head Internship
April 2022 - May 2022
Data Analyst
Feb 2022 - March 2022
- Designed and implemented predictive models and dynamic dashboards using Python, Power BI, and Tableau, enhancing decision making.
- Developed a machine learning pipeline to automate data processing and visualization tasks, reducing manual effort by 40%.
- Managed and mentored a team of 10-15, drove client acquisition and retention efforts, expanding the client base by 30%.
- Enhanced internship program quality, introducing complex tasks which offered scholarships to best-performing interns.
- Data Analyst (Feb 2022 - March 2022): Led a team of 8 in harnessing advanced analytics to optimize marketing strategies and improve campaign effectiveness.
- Utilized Python and visualization tools to analyze, build models, and run custom ad campaigns, resulting in a 25% increase in ROI.
- Delivered compelling data visualizations and presentations to senior management, effectively communicating strategic insights that guided critical business decisions.
WT
Wheel and Time
Data Analyst
June 2021 - July 2021
WT
Wheel and Time
Data Analyst
June 2021 - July 2021
- Developed quantitative analysis models to support key operation decisions, optimizing processes and improving efficiency.
- Compiled monthly logistics financial reports, ensuring accurate data presentation and aiding in financial decision making.
Projects
Showcasing academic research, publications, and production-ready systems.
Live Formula 1 Race Strategy and Analytics Platform
AI / ML Engineer
June 2026 - Present
Los Angeles, CA
- Architected a CrewAI based multi-agent system for real-time Formula 1 race strategy generation, utilizing LangGraph for state orchestration within a deterministic 10Hz telemetry processing loop to enable collaborative LLM-driven decision making for optimal pit stop timing and tire compound selection strategies.
- Developed a hybrid anomaly detection framework combining adaptive statistical thresholds, including z-score based pacing anomaly detection and sensor fusion techniques, with Turbovec powered vector similarity search for rapid identification of incipient mechanical faults and performance degradation patterns in high-dimensional streaming telemetry data.
- Engineered WebSocket telemetry ingestion with sub-50 millisecond latency, a custom positioning algorithm for accurate SVG track visualization, and an interactive driver drill-down analytics dashboard delivering real-time actionable insights including lap time delta analysis and strategic recommendations.
Live Pothole Detection System
Project Leader, ML Engineer, Website Developer
June 2022 - June 2023
Mumbai, India
- Developed a software leveraging live video feeds from traffic signal cameras to detect potholes on road and report them to the concerned authority of respective area along with details such as location and photos of reported potholes; potholes can also be reported by users of application, and relevant details would be forwarded to respective authorities.
- Engineered machine learning model which was an integral part in project leveraging Python and CNN model, object detection, etc., trained CNN model and integrated it in web and mobile applications developed for live monitoring and for user reporting.
- Programmed website for showing live statistics of reports and results.
Research Driven Big Data Hybrid Recommendation System
ML Engineer
Jan 2024 - March 2024
Los Angeles, CA
- Designed and Experimented with a Hybrid Recommendation System, initially combining item-based collaborative filtering with XGBoost. The hybrid approach was tested extensively to explore its potential in improving prediction accuracy for Yelp review ratings.
- Conducted Rigorous Research and Comparative Analysis of various modeling techniques, including collaborative filtering and model-based methods, to determine the optimal approach for handling diverse user-business interactions.
- Finalized a High-Precision Model-Based Recommendation System after extensive experimentation, leveraging Spark RDDs and XGBoost. This model, enriched with advanced features such as user engagement metrics and business attributes, achieved an industry-leading RMSE of 0.9784.
- Enhanced Computational Efficiency and Scalability by utilizing distributed processing with Spark, reducing system latency by 30%, and enabling real-time predictions on large-scale datasets.
Custom Query Language and Scalable Database System
Database Engineer
Jan 2024 - March 2024
Los Angeles, CA
- Designed and implemented a scalable database engine that efficiently supports both relational and NoSQL models, with chunked data loading and processing, optimized for large datasets without memory overflow.
- Developed a custom query language processor enabling SQL-like operations through a syntax parser using regex and mapping techniques, enhancing flexibility in data retrieval and manipulation.
- Engineered efficient data retrieval and manipulation using chunked data processing with custom sorting and filtering functions, optimized for large scale real world datasets.
Academic Roots
The foundation of my analytical and engineering capabilities.
University of Southern California
Master's, Applied Data Science
August 2023 - May 2025
Los Angeles, CA
Thadomal Shahani Engineering College
Bachelor's, Computer Engineering
June 2019 - June 2023
Mumbai, India
Global Certifications
Accredited by industry leaders in AI, Data Science, and Systems Architecture.
Generative AI for Data Scientists Specialization
IBM / Coursera
Sep 2024
Business Analysis by Microsoft
Microsoft / LinkedIn
Sep 2024
Deep Learning Specialization
DeepLearning.AI
2024
Database Engineering
Kaggle
2023
Python for Data Science
Kaggle
July 2023
Advanced C++ Certification
HackerRank (3 Star)
2023
Python Programming
Udemy
May 2021