Software Developer · AI / ML

Ashraful Islam

MS CS (Intelligent Systems) @ UT Dallas
Prev AI Engineering & Software Developer Intern @ Gainwell Technologies
Building intelligent systems, full-stack platforms, and AI-driven solutions

About

"AI" might not be my middle name — but it is my initials.

I'm a developing software systems engineer / full-stack developer with a focus on AI / ML / DS. I've gained experience across responsive front-end UIs and libraries / frameworks, backend services, databases, and APIs, through my internships, freelance / consultancy work, coursework, projects, and multiple large-scale hackathons.

I seek opportunities to contribute my talents in environments where I can make a difference, bringing excellent leadership, service, and innovative IT solutions to benefit communities and advance public well-being.

Check out my experience, projects, and research below.

Education

Master of Science in Computer Science

The University of Texas at Dallas · Intelligent Systems

Aug 2025 – May 2027
  • Coursework: Machine Learning, Artificial Intelligence, Big Data Analytics, Database Design, Design of Algorithms

Bachelor of Science in Computer Science

The University of Texas at Dallas · Magna Cum Laude · Honors, Dean's List · GPA 3.95 / 4.00

Aug 2023 – May 2026
  • Coursework: Data Structures & Algorithms, Software Engineering, Operating Systems, Computer Architecture, Linux

Experience

Artificial Intelligence Engineering Intern

Gainwell Technologies LLC · Irving, TX

May 2026 – Aug 2026
  • Led a 33-member cross-functional team to develop an internal AI release notes assistant integrating Azure DevOps, GitHub, and SharePoint; pitched to executives and won 1st in the intern capstone deployment competition
  • Automated release note generation and knowledge retrieval, delivering a projected $2M+ annual labor savings by saving 20 hrs per release across 100 monthly production releases, with estimated AI costs <1% of savings
  • Improved SNAP Benefits Integrity (FWA) ML models on Databricks, increasing F1 score by 5.4% on sparse data through parameter optimization (+5.3%) and training a new national recipient error model (+0.1%)
  • Compared national vs. state (OK, FL, KY) SNAP models' performance to select the optimal model (81% F1)
  • Built data-enrichment processes to enhance 2,000+ attribute descriptions for a Text-to-SQL PBA data agent
  • Designed a pharmacist-facing UI for MediSpan-FDB Rx data labeling/matching, leveraging CPhT expertise

Software Developer Intern

Gainwell Technologies LLC · Irving, TX

May 2025 – Aug 2025
  • Led an 8-member cross-functional team to design and deploy 30+ Power BI dashboards integrating SAP, SOC 1, and ITGC datasets, enabling automated risk and compliance analytics and executive decision support
  • Presented to executives and won 1st place against 11 teams in company-wide innovation competition
  • Executed claims validation tests, T-SQL query automation scripts, documentation, and C#.NET/ADO workflow optimization with senior developers and business analysts to improve TX/OH healthcare enterprise PBM systems
  • Authored a 50-page technical manual adopted as the team's standard onboarding documentation

Software Engineer

Headstarter AI · Remote

Jul 2024 – Sep 2024
  • Led a 4-engineer agile team to deliver 4 full-stack AI applications in weekly production sprints and code reviews
  • Developed AI systems including RAG RateMyProfessor, LLM-powered flashcard generator, and hotel chatbot with Gemini
  • Integrated authentication, payments, and cloud deployment using Clerk, Stripe, Firebase DB, and Vercel
  • Demoed fellowship projects for company review/feedback on professional development and performance scores

Shift Supervisor Rx, Certified Pharmacy Technician (CPhT)

CVS Pharmacy · Frisco, TX

Sep 2022 – Present
  • Cross-trained as shift manager and CPhT; exceeded performance targets with 150+ membership conversions
  • Support pharmacy operations serving 100–250 patients per shift while maintaining HIPAA data compliance

Skills

Languages

Python Java C# JavaScript TypeScript SQL C/C++ HTML (Tailwind) CSS Bash s(CASP)/Prolog

Frameworks & Libraries

Next.js React.js Nuxt.js Vue.js ASP.NET PyTorch NumPy Pandas Apache Spark Kafka

Databases

Postgres MySQL MS SQL Server Firebase MongoDB Hadoop (HDFS) / Hive HBase Cassandra

Dev / AI Tools

Git / GitHub Linux / WSL Azure DevOps GCP AWS (AIF-C01) Vercel Databricks Power BI RAG Vector DB Vertex AI Google ADK Visual Studio

Projects

Across AI / ML systems, full-stack web apps, and big-data pipelines — spanning hackathon wins, client work, coursework, and personal explorations.

AI / ML

MedVisor — Multi-Agent Clinical Diagnostics

Architected a parallel multi-agent AI diagnostic system with 3 concurrent specialist agents (ClinicalScribe, RadiologyAnalyst, RecordsAnalyst) achieving 3× faster processing. Integrated the WHO ICD-11 REST API and openFDA (140K+ drug labels) for hallucination-free, real-time diagnosis validation and FDA-labeled treatment retrieval across 55K+ medical codes. Deployed a multimodal MCP server on Cloud Run with audio, image, and EMR input support and structured JSON output.

Python Google ADK Gemini 2.5 Flash Vertex AI Cloud Run

Autis(CASP) — 1st Place HackReason

Won 1st Place at HackReason by developing a user-friendly, logic-based autism screening AI chatbot. Led design on a research-backed screening tool to evaluate input on 7 core symptoms and DSM-5 criteria. Integrated s(CASP) with Gemini-LLM and Python to deliver explainable, medically grounded responses.

Next.js React.js s(CASP) Prolog Python Gemini AI

ImmerseAI — HackAI 2026

Built a personalized learning-path generator using a 3-agent Google ADK pipeline (Gemini 2.5 Flash) for knowledge-gap analysis, content discovery, and prerequisite validation. Integrated s(CASP) logical reasoning over the YouTube Data API to vet educational resources and synthesize prerequisite-aware roadmaps with explainable justifications. Deployed a FastAPI + FastMCP backend on Cloud Run with MongoDB Atlas state persistence and a React + Vite client for interactive roadmap exploration.

Python Google ADK Gemini 2.5 Flash FastAPI FastMCP s(CASP) React + Vite MongoDB Atlas Cloud Run

PawPal+ AI — Applied AI System Scheduler

Architected a Retrieval-Augmented Generation (RAG) system grounding LLM-generated pet-care schedules on a 33-document curated knowledge base; ChromaDB persistent vector store with 384-d MiniLM embeddings serves top-5 cosine-similarity retrieval. Engineered a hybrid recommender with similarity-weighted ranker and a 3-tier LLM cascade (Gemini → Claude → deterministic template) for zero-key fallback. Designed a full-recurrence busy-block calendar, medical-claim guardrails, and structured JSON audit logging; verified with 107 unit tests and a 6-scenario reliability harness, deployed on Streamlit Cloud.

Python Streamlit ChromaDB Sentence-Transformers Gemini API RAG

MediCASP

Developed an AI medical screening chatbot using Gemini AI and Next.js to provide 24/7 symptom assessment. Engineered a logic engine with s(CASP) to automate scientific validation for 8 medical conditions based on patient data. Designed a modern user interface with Next.js and Clerk API to provide user-friendly medical screening.

Next.js React.js s(CASP) Prolog Python Gemini AI Clerk

AI-Powered Applications Suite

Developed an AI flashcard generator for adaptive learning, a hotel support chatbot with natural language processing, a RAG-based RateMyProfessor for intelligent course recommendations, and a cloud-based inventory management system. Integrated Gemini AI, Clerk, Stripe, and Firebase into modern full-stack deployments.

React.js Next.js Gemini AI Firebase Clerk Stripe

Full-Stack Web Apps

Client Masjid Website — 1st Place Khidmah Consultancy

Won 1st Place at the Khidmah Consultancy competition; shipped a full-stack platform slated to replace the client's legacy site. Architected a cascading 3-step data layer across 9 Prisma models, enabling zero-downtime rendering of prayer times, events, and Hijri dates against multiple upstream APIs. Engineered a recurrence-aware event calendar, HMAC-signed admin email approval workflows, multi-tier rate-limited forms with honeypot anti-spam, and Hijri/Gregorian sync via dual-API fallback.

Client Project Next.js React.js TypeScript Prisma Supabase Resend Vercel

Jewelry E-Commerce

Led development on a scalable, full-stack e-commerce platform for a luxury jewelry business featuring product collections, shopping cart, and a secure multi-option checkout flow (Stripe, Zelle, and bank wire). Optimized performance by 28% through image tuning and reusable UI components. Implemented backend routing and database schemas for catalog, cart, and transaction workflows.

Client Project

T.R. Hoover CDC Data Management System

Built a Nuxt/Vue.js full-stack internal data management app with a central database to support grant reporting. Designed a scalable Prisma schema for member PII (100–300+ per event), attendance, and volunteer/admin data. Developed an intuitive UI for quick member check-in and data retrieval; demoed to sponsor with perfect rating.

Nuxt.js Vue.js Prisma SQLite

AeroScope

Developed a flight tracker UI using Aviation Stack flight data and React for DFW International Airport. Tracks and filters 1800+ arriving and departing flights by 5 different categories. Implemented user authentication using Clerk for users to access the webpage and live flight information.

React.js HTML CSS Clerk API Aviation Stack API

FitTrack

Built a MERN web application to plan and track fitness goals and assist with dietary needs. Integrated a Llama LLM chatbot for workout recommendations and nutritional information. Created a React.js UI to retrieve and display user sign-up, sign-in, and fitness data.

MongoDB Express.js React.js Node.js JavaScript Llama LLM

Research

Formal methods for safety-critical avionics and applied ML for immersive media.

Formal Verification in Integrated Avionics Systems — AerosCASP

AIAA Region IV Student Conference · The University of Texas at Dallas · Mar 27–29, 2026

William T. Doan, Ashraful Islam, Viknesh Venkatachalam, Mahd Malik

Jan 2026 – Mar 2026
  • Co-authored an AIAA Region IV Student Conference paper proposing a “Correctness by Construction” framework that shifts DO-178C avionics verification from post-implementation testing to formal proof during the requirements phase of the V-Model
  • Modeled continuous flight dynamics (Minimum Safe Altitude / GPWS logic) in s(CASP) with CLP(Q) rational constraints, mathematically proving safety properties over infinite state spaces without finite grounding
  • Automated the Requirements Traceability Matrix (RTM) via s(CASP) justification trees and #pred natural-language directives, and used Dynamic Consistency Checking (DCC) to prune ARINC 653 IMA scheduling search spaces during model generation
  • Built AerosCASP, the working implementation of the published framework, encoding the paper's avionics models and proofs as executable s(CASP) programs
s(CASP) Answer Set Programming CLP(Q) Prolog DO-178C ARINC 653 LaTeX

360° VR Attention ML FoV Prediction System

Undergraduate ML research · The University of Texas at Dallas

Aug 2025 – Dec 2025
  • Designed and trained a CNN-based classification model on a 144-class output space to predict user head orientation and reduce theoretical VR streaming bandwidth through data-driven field-of-view prediction
  • Fused visual, motion, and audio saliency heatmaps from equirectangular projections into per-frame FoV predictions, selecting a tiled subregion to stream
  • Engineered a large-scale data pipeline processing 40B+ spatiotemporal samples for model training and validation
  • Evaluated performance on top-1 metrics, achieving 50.6% validation accuracy and 31% on real unseen data
PyTorch CNN Python Saliency Fusion ML Research