Product · AI / ML
Portfolio
Scroll-driven CSS transitions and Mimir, a retrieval-only chat that answers with real passages and links.
aqeed
safwaan
Building robust AI systems,
from research to production.
Open to full-time SWE / ML roles · 2027
Mimirfrom this site
Hi, I'm Mimir. Ask me about Aqeed's work, projects, skills or what he's looking for. I only answer from what's on this site.
Limited to this site: answers come straight from its content.
$ cd /about
about me
Building robust AI systems, from research to production.
AI engineer at GPU IQ and Computer Science master's student at Rutgers, graduating May 2027. Day-to-day tools: Python, TypeScript, PyTorch, Next.js, PostgreSQL and the Claude API. Interested in LLM applications, retrieval and evaluation, data pipelines, and machine learning systems that hold up in production. Seeking a full-time ML, AI or software engineering role starting summer 2027.
AI Engineer Intern @ GPU IQ
in_my_element.gifBy the numbers
- 1+Years of experienceGPU IQ · Medisaya
- 7ProjectsAll listed in Work
- 63Tech skillsAcross 5 domains
Roles
AI / ML Engineer
RAG evaluation, LLM agents and token-budget-aware prompting, running in production.
SaaS Architect
Multi-tenant platforms: row-level isolation, Stripe billing and onboarding.
Product Developer
Full-stack SWE — Next.js and React through Django, Flask and PostgreSQL.
Applied ML Researcher
Benchmarking CNNs, Transformers and SSMs, and distilling models for the edge.
What I'm building right now
Interconnection-queue intelligence
A live pipeline across all six US ISOs — scraping, county-level geocoding and scoring for 16,291 active projects.
- Next.js
- TypeScript
- PostgreSQL
JUNO, a public AI chat widget
Claude-powered, answering from live production signal data and routing visitors by intent to a booking or a self-serve signup.
- Claude API
- Next.js
- TypeScript
RAG evaluation harness
Per-case chunk inspection and pass-rate diffing in server components — 78.25% across 400 logged evals.
- Next.js
- PostgreSQL
- Claude API
Multi-tenant SaaS foundations
Tenant isolation across 107 RLS-protected tables, with Stripe billing, onboarding and a meeting-bot pipeline on top.
- PostgreSQL
- Stripe
- Next.js
Focus areasRAG & evalsLLM agentsMulti-tenant SaaSData pipelinesEdge ML
What I can build — and with what
- 01
LLM apps & agents
Chat agents, RAG pipelines and eval harnesses
- Claude API
- vLLM
- Hugging Face
- PyTorch
- 02
Multi-tenant SaaS
Isolation, billing, onboarding and auth
- Next.js
- TypeScript
- PostgreSQL
- Stripe
- 03
Data & ML pipelines
Ingestion, geocoding, training and evaluation loops
- Python
- Pandas
- NumPy
- Scikit-learn
- 04
Vision & edge ML
Model benchmarking, distillation and edge inference
- PyTorch
- TensorFlow
- Keras
- OpenCV
- 05
Full-stack products
UI to API to database, end to end
- React
- Next.js
- Django REST Framework
- Flask
- MongoDB
Experience
AI Engineer Intern · GPU IQMay 2026 – Present
- Next.js
- TypeScript
- PostgreSQL
- Claude API
- Engineered a live interconnection queue pipeline across all 6 US ISOs via dynamic URL scraping and Census TIGERweb county-centroid geocoding (3,235 counties, 95.9% coverage), ingesting 16,291 active projects into PostgreSQL through a monthly cron to automate upsert, proximity-ring scoring, and account rescoring.
- Redesigned a RAG evaluation harness's review workflow into Next.js server components with per-case retrieved-chunk inspection and pass-rate diffing, surfacing a 78.25% pass rate across 400 logged eval results while eliminating client loading states via server-side DB fetch.
- Hardened the AI Teammates agent panel by engineering token-budget-aware system prompts that resolved response truncation (60% to 100% completion), then built a guided 4-step account workflow with Salesforce-synced deal-stage tracking and a 5-stage AI pipeline auto-runner eliminating manual agent handoffs.
- Built JUNO, a public-facing AI chat widget (Claude Opus 4.8, adaptive extended thinking) querying live production signal data through a 9-source trust allowlist and 5-category GPU/AI taxonomy filter, routing visitors via a weighted intent-scoring model (score ≥ 80) to meeting booking vs. self-serve signup.
- Identified and fixed two tenant-isolation vulnerabilities enabling cross-user data exposure, led a dependency-hardening pass cutting vulnerabilities from 25 to 8, and migrated a 288-file multi-tenant data-isolation layer to async request-scoped context for a Next.js 16 / Turbopack upgrade with zero regressions across 107 RLS-protected tables.
- Shipped customer-lifecycle infrastructure spanning self-serve billing and onboarding (Stripe Checkout, 3 metered tiers, 14-day trial, 2-step funnel) through post-signup engagement via a calendar-integrated AI meeting-bot pipeline syncing 5 providers to schedule Recall.ai bots with Claude Sonnet 4.6 transcription.
Full-Stack Software Developer · MedisayaFeb 2025 – Aug 2025
- Django
- vLLM
- TypeScript
- Git
- Integrated cloud software services to engineer a low-latency notification pipeline using Expo and Firebase within Agile workflows, delivering real-time health alerts with 99.9% uptime to boost medication adherence by 15%.
- Implemented a 'Friends' social graph using Django REST APIs, enforcing role-based access control (RBAC) to securely manage patient-caregiver data sharing permissions.
- Developed full-stack data pipelines using Django, React, and CI/CD, implementing automated metric normalization to standardize diverse medical inputs and reduce database storage overhead by 30%.
- Automated CRUD operations via GenAI and LLM workflows, NLP, and tailored test case learnings, reducing administrative time by 40%.
Education
Rutgers UniversityAug 2025 – May 2027
Relevant coursework
- Machine Learning II
- Introduction to Artificial Intelligence
- Topics in Artificial Intelligence
- Brain Inspired Computing
- Massive Data Mining
- Database Systems for Data Science
- Data Structures and Algorithms
- Topics of Computers in Biomedicine
- Internet Services and Design
BRAC UniversityFeb 2021 – Feb 2025
Relevant coursework
- Artificial Intelligence
- Machine Learning
- Robotics
- Data Structures
- Algorithms
- Database Systems
- Operating Systems
- Computer Networks
- Computer Architecture
- Compiler Design
- Computer Graphics
- Software Engineering
- Systems Analysis and Design
- Microprocessors and Microcontrollers
- Computer Interfacing
- Discrete Mathematics
- Ethical Hacking
Sunnydale School2018 – 2020
8 O-Level subjects, 4 A-Level subjects
Certificates
AnthropicOct 2026
IBMMar – Jun 2026
Amazon Web ServicesJan 2026
$ cd /work
work
AI / ML
Autonomous Localization & Path Planning
Heuristic search for grid localization: a learned heuristic expands about 95% fewer nodes than optimal search.
Product · Systems
BuyMe Auction Site
A live Java auction site: bid by hand or set a secret auto-bid limit, across 1,740 real listings.
AI / ML
Credit Card Fraud Detection
Five fraud models compared on 1M transactions, with precision-recall curves and a cost-based threshold.
Research · AI / ML · Systems
CV Image Classification for Edge Wildfire Detection
CNNs vs Transformers vs SSMs for edge inference, distilled into a small SSM: 65% to 85.3% accuracy.
$ cd /skills
tech skills
$ ls
Languages
PythonJavaCC++JavaScriptTypeScriptSwiftPHPSQLHTMLCSS
$ ls
AI & ML
PyTorchTensorFlowKerasHugging FaceScikit-learnOpenCVNumPyPandasMatplotlibJupytervLLMDynamoClaude APIMCPLLMsTransformersNLPRAGAI agentsPrompt engineeringCNNsState space modelsKnowledge distillationModel evaluation
$ ls
Web & Mobile
ReactNext.jsExpoAndroid SDKFirebase
$ ls
Backend & Data
Django REST FrameworkFlaskJava Servlets & JSPApache TomcatPostgreSQLMySQLTiDBSQLiteMongoDBStripeSalesforceRecall.aiWeb scrapingMulti-tenancy & row-level security
$ ls
DevOps, Cloud & Security
GitDockerAWSVercelCI/CDTest AutomationBurp SuiteRBACAgile
$ cd /contact
get in touch
Send a message directly, or reach me another way:
$ mail -s
Send a message
$ contact
- GitHubgithub.com/aqeed01
- LinkedInlinkedin.com/in/aqeed01