teamir.exe—Professional Runtime Environment|Status: ACTIVE|Role: Software Engineer|Graduation: Aug 2026_
> cat education.txt
B.S Computer Science
Georgia State University
Expected Aug 2026
Major GPA: 4.0/4.3
Relevant Coursework:
AlgorithmsData StructuresSoftware DevelopmentMachine LearningDatabase SystemsMobile App DevelopmentSystems ProgrammingProgramming Language ConceptsComputer ArchitectureProbability & StatisticsCalculus IILinear Algebra
Honors & Awards:
President's ListExcellence ScholarshipEmoryHacks '25 Winner
Organizations:
Programming Club at GSUNational Society of Black EngineersAfrican Student Association
High School Diploma
Dekalb Early College Academy
May 2022
GPA: 4.2
Relevant Coursework:
60 Early College Credits at Georgia State University
Honors & Awards:
President's ListPrincipal's ListStudent of the Month
> cat experience.txt
Incoming Software Engineer Intern
Cargill
Summer 2026
Undergraduate Researcher
CHAI Lab and Morse Studio
Sept 2025 - Present
Joining two research labs, excited to share what we work on!
AI / Document Intelligence Extern
Outamation
Aug 2025 - Present
- Ran retrieval experiments across embedding models (MiniLM/E5/BGE) and retriever settings (top-k, similarity thresholding, optional reranking), selecting a configuration that improved document-grounded QA on mortgage fee worksheets.
- Implemented hybrid ingestion (PyMuPDF text extraction + Tesseract OCR fallback) and validated OCR triggers using text-length heuristics to avoid unnecessary OCR on clean digital PDFs.
- Optimized chunking parameters (window size, overlap) to preserve contract clauses and fee tables, improving retrieval coherence and reducing context fragmentation.
- Added metadata tagging + filtering (doc_type, source_file, page/chunk IDs) to route queries and reduce noise during multi-PDF "blob" retrieval.
- Measured pipeline outputs (confidence, retrieved sources, latency, chunks-used) on a query test set and iteratively refined prompts and retrieval logic based on observed failure modes.
- Shipped an interactive Gradio chatbot wired to the full RAG backend (LlamaIndex + FAISS + llama.cpp LLMs) to demo and validate the system end-to-end.
Undergraduate Teaching Assistant - Calculus & Statistics
Georgia State University
Jan 2024 - May 2025
- Planned, organized, and led weekly peer-assisted study sessions for Calculus and Statistics courses, independently designing session structure, problem sets, and instructional strategies to reinforce core mathematical concepts and exam readiness.
- Proactively marketed and promoted sessions through campus outreach and peer networks, driving consistent attendance and engagement without direct faculty oversight.
- Facilitated large-group problem-solving discussions, breaking down complex concepts into clear, intuitive explanations that enabled students of varying backgrounds to reach understanding and apply techniques independently.
- Mentored students in independent learning strategies, emphasizing analytical thinking, self-debugging, and effective study methods rather than dependency on instructors or tutors.
- Built strong cross-year peer relationships, collaborating with upperclassmen and repeat attendees to foster a supportive learning environment and sustained academic progression.
- Earned 100% satisfaction ratings in anonymous student feedback surveys, with students citing leadership, clarity of instruction, and measurable improvement tied directly to session participation.
Tutoring Data & Systems Clerk
Saturday Academy - RAOC
Aug 2025 - Present
- Managed and organized student performance and attendance data to track tutoring progress, identify trends, and support data-driven instructional decisions across multiple students.
- Designed and maintained streamlined data-tracking processes for recording outcomes, improving consistency, accuracy, and ease of analysis for tutors and administrators.
- Supported mentorship and tutoring initiatives by working directly with younger students, building strong relationships that encouraged engagement, confidence, and long-term academic growth.
- Collaborated with tutors and staff to translate performance data into actionable insights, helping tailor support strategies to individual student needs.
- Planning and developing a web-based platform to centralize student records, streamline reporting workflows, and improve accessibility and transparency for stakeholders
Founder
TMD Detailz
Jan 2024 - Dec 2025
Founded and operated a successful detailing business in Stone Mountain, Georgia. Managed all aspects of business operations including client relations, service delivery, and business development.
> ls projects/
CORE SERVICE
XP Lab (EmoryHacks '25 Winner)
STABLE
Stack: Python, FastAPI, React.js, PostgreSQL (Supabase), WebSockets, Whisper API
Focus: AI-powered lecture intelligence platform for professors and students
- Built a real-time, two-sided education platform that improves lecture quality for professors while simultaneously increasing student engagement through AI-driven feedback, analytics, and gamification.
- Designed and implemented a live lecture pipeline using FastAPI, WebSockets, and OpenAI Whisper to stream audio, transcribe lectures in real time, and surface pacing, engagement, and confusion insights during and after class.
- Engineered a gamified student experience (points, streaks, ranks, badges, leaderboards) with React, Tailwind CSS, and Vite to motivate participation, reinforce consistency, and visualize learning progression.
- Developed a scalable backend architecture with Supabase (PostgreSQL), role-based authentication, and real-time session management to support concurrent classes, students, and professors.
- Integrated AI-assisted question generation and hybrid workflows that allow professors to combine manual input with AI suggestions, enabling rapid in-class assessments without disrupting lecture flow.
Architecture: Scalable backend architecture using Supabase (PostgreSQL), role-based authentication, and real-time audio streaming
Decisions: Engineered gamified learner experience with points, streaks, badges, and leaderboards that increased student engagement
PRODUCTION SYSTEM
JobHunter.AI
STABLE
Stack: Python, Flask, React.js, PostgreSQL (Supabase), OpenAI, Google Cloud, Azure, Matplotlib, Plotly
Focus: Automated job application tracking system
- Designed and built an end-to-end automated job application tracking system that removes the need for manual data entry by continuously ingesting job-related emails from Gmail and Outlook.
- Implemented OAuth-secured email integrations with Google and Microsoft APIs to safely access user inboxes, manage token lifecycles, and support multiple connected email accounts per user.
- Developed an AI-powered email intelligence pipeline using OpenAI that classifies incoming messages, extracts structured fields (company, role, location, status), and updates application records automatically.
- Engineered a Flask-based backend architecture that orchestrates email ingestion, AI processing, background synchronization, and database persistence in a reliable and maintainable manner.
- Built a responsive React dashboard with search, filtering, and status management that allows users to view and control their entire job search across multiple inboxes from a single interface.
Architecture: Flask backend handles email ingestion, parsing, and database updates with continuous data-processing pipeline
Decisions: Created interactive visual analytics (Matplotlib, Plotly) to help users visualize application progress and trends
PRODUCTION SYSTEM
Document_Intelligence.RAG
STABLE
Stack: Python, LlamaIndex, Hugging Face, Mistral-7B, Phi-2, FAISS, Tesseract OCR, Gradio
Focus: End-to-end document intelligence system for mortgage-related PDFs, combining OCR, metadata-aware retrieval, and open-source LLMs to enable accurate, source-grounded question answering.
- Designed and implemented a full Retrieval-Augmented Generation (RAG) pipeline capable of processing both digital and scanned mortgage documents.
- Built a selective OCR fallback using Tesseract to extract text from low-quality or scanned PDFs while avoiding unnecessary OCR on digital documents.
- Engineered a chunking and metadata tagging strategy to preserve legal clauses and fee sections, enabling targeted and semantically accurate retrieval.
- Integrated BGE embeddings with a FAISS-backed vector index via LlamaIndex to support fast semantic search over document chunks.
- Implemented a mixed-model architecture using Phi-2 for query routing and document classification, and Mistral-7B for final answer generation.
- Developed a RAG prompt strategy that constrains LLM responses to retrieved context, returning answers with citations and confidence scores.
- Built an interactive Gradio chatbot interface to support document upload, chat history, and real-time question answering.
- Evaluated routing accuracy, retrieval consistency, confidence scoring, and end-to-end stability across multiple mortgage document types.
Architecture: Modular RAG architecture with clear separation between document ingestion (OCR & parsing), semantic indexing (chunking, embeddings, vector search), and reasoning (LLM-based answer generation), designed to mirror production document AI systems.
Decisions: Mixed-model design chosen to balance latency and reasoning quality; lightweight embeddings selected for efficient indexing; metadata-aware retrieval used to improve precision; CPU-only Colab environment prioritized stability and reproducibility over raw inference speed.
Focus: End-to-end email spam detection using traditional NLP techniques with emphasis on reproducibility, validation discipline, and clean ML pipeline design.
- Designed and implemented a complete machine learning pipeline to classify email messages as spam or ham using TF-IDF feature extraction and a linear Support Vector Machine.
- Ingested and normalized multiple training datasets with differing schemas, unifying labels and text fields into a consistent representation.
- Engineered a text preprocessing workflow including stop-word removal, lowercasing, and minimum document frequency filtering to reduce noise in high-dimensional feature space.
- Trained and evaluated a baseline LinearSVC model using a stratified train/validation split, reporting accuracy, precision, recall, F1 score, and full classification metrics.
- Performed hyperparameter tuning with 5-fold cross-validation using GridSearchCV, optimizing for F1 score to balance false positives and false negatives.
- Re-trained the final model on all available labeled data and generated predictions for an unseen test dataset, persisting results to a structured output artifact.
- Refactored exploratory notebook code into a standalone, reproducible Python script with clear separation between data, source code, and outputs.
Architecture: Script-based ML pipeline with explicit separation of data ingestion, preprocessing, feature extraction, model training, validation, and inference; exploratory development in Jupyter/Colab with a production-style execution script.
Decisions: TF-IDF selected for interpretability and efficiency on sparse text data; Linear SVM chosen for strong performance in high-dimensional NLP tasks; validation-based evaluation used to prevent data leakage due to unavailable test labels; notebook retained for exploration with a cleaned script for reproducible execution.
INFRASTRUCTURE
Employee Management System
ARCHIVED
Stack: Java, MySQL, JDBC, SQL, Console-Based Application Architecture
Focus: Secure employee data management with role-based access control and payroll reporting
- Designed and implemented a Java-based employee management system that provides secure, role-based access for HR administrators and employees to manage and view sensitive employee data.
- Built a role-based authentication and authorization layer that enforces full CRUD privileges for HR administrators while restricting general employees to read-only access.
- Developed a robust relational database schema in MySQL with normalized tables, foreign key constraints, and many-to-many relationships to model employees, divisions, job titles, and payroll records accurately.
- Implemented employee search and management workflows that support queries by employee ID, name, SSN, and date of birth, with real-time updates persisted safely to the database.
- Engineered payroll and reporting functionality that generates monthly pay summaries by job title and division, employee-specific pay histories, and hire-date range reports.
- Used prepared statements throughout the data access layer to prevent SQL injection and ensure secure interaction with the database.
- Structured the application with a clear separation of concerns (authentication services, business logic, repositories, and database access) to improve maintainability and extensibility.
Architecture: Layered architecture with separation of concerns: authentication services, business logic, repositories, and database access
Decisions: Prepared statements throughout data access layer for security, normalized database schema with foreign key constraints
PRODUCTION SYSTEM
TeamirTeshome.com
STABLE
Stack: Next.js, TypeScript, React, Framer Motion, Tailwind CSS, Next.js Image Optimization
Focus: Dual-mode portfolio website with OS-themed professional interface and personal reflection layer
- Designed and implemented a bootloader-style landing page that presents two distinct modes (teamir.exe and teamir.raw) with terminal-inspired UI and smooth navigation transitions.
- Built a professional OS-themed interface (teamir.exe) that displays projects, education, and technical experience using system-inspired components with status indicators and process management metaphors.
- Developed a personal reflection layer (teamir.raw) featuring a horizontal scrolling timeline, paginated journal entries, media gallery, and curated personal metadata with warm, low-contrast dark aesthetic.
- Engineered a centralized theme system with shared color palettes and motion presets to ensure visual consistency across landing, professional, and personal pages.
- Implemented a motion-only tricolor accent system that reveals Ethiopian identity through subtle animated color sweeps on interactive elements while maintaining a minimal monochrome resting state.
- Created reusable animation components using Framer Motion with useAnimation hooks to enable persistent hover effects that play tricolor sequences and maintain enhanced states during interaction.
- Designed responsive layouts with mobile-first approach using Tailwind CSS grid and flexbox systems that adapt seamlessly across desktop, tablet, and mobile viewports.
- Integrated Next.js Image optimization and API routes to serve dynamic image galleries with automatic format conversion and lazy loading for improved performance.
- Architected a modular component structure with shared HoverableCard components, theme constants, and motion presets to enable consistent styling and behavior across all sections.
Architecture: Next.js App Router with client-side routing; shared theme and motion configuration in lib/ directory; modular page components (landing, professional, personal) with reusable UI primitives; Framer Motion for declarative animations; Tailwind CSS for utility-first styling.
Decisions: OS metaphor chosen to differentiate from traditional portfolios and create memorable user experience; dual-mode design separates professional and personal content while maintaining system cohesion; tricolor accent system adds cultural identity without overwhelming minimal aesthetic; TypeScript selected for type safety across complex component hierarchies; Framer Motion used for performant declarative animations over imperative DOM manipulation.
HARDWARE
HearSpace
EXPERIMENTAL
Stack: Python, Computer Vision, Embedded Systems, ESP32-CAM, Deep Learning Models, Spatial Audio
Focus: Assistive navigation through real-time vision-to-audio spatial mapping
- Designing an assistive navigation system that integrates embedded hardware, computer vision, and audio feedback to translate visual spatial information into intuitive sound-based cues.
- Developing a real-time perception pipeline using camera input and deep learning models to detect obstacles and estimate depth for navigation assistance.
- Integrating edge hardware components (ESP32-CAM and embedded camera modules) with backend processing logic to support low-latency data capture and transmission.
- Exploring spatial audio mapping techniques to convert object position and distance into directional audio cues that improve user situational awareness.
- Engineering the system incrementally with a modular architecture, allowing vision models, audio logic, and hardware components to evolve independently during development.
Architecture: Modular architecture with separate components for vision processing, audio mapping, and hardware integration
Decisions: Incremental development approach allows independent evolution of vision models, audio logic, and hardware components
> cat certs
Intermediate Technical Interview Prep
COMPLETE
AI Document Intelligence Extern
COMPLETE
Microsoft Azure AI Fundamentals
IN PROGRESS
Grokking the System Design Interview
IN PROGRESS
Github Certified Developer
IN PROGRESS
> cat skills.txt
Languages
- Python
- SQL
- C
- JavaScript
- HTML
- CSS
Frameworks/Libraries
- Flask
- FastAPI
- React.js
- Librosa
- PyTorch
- NumPy
- Pandas
- Matplotlib
Tools/Databases
- Git/Github
- Supabase
- MySQL
- PostgreSQL
- Google Cloud
- Azure
- OpenAI
> cat interests.txt
- • AI + Systems Engineering
- • Full-Stack & ML Development
- • Applied Machine Learning
- • Backend Architecture & Performance
- • Systems Programming
> cat trajectory.txt
Right now, I’m primarily locked in on software engineering, so I’m investing heavily in core SWE skills through things like a systems design certification, an AWS/cloud certification, and an upcoming SWE internship. At the same time, I’m intentionally building strong AI/ML fundamentals by joining two research labs and working on ML-focused projects, with the long-term goal of specializing in machine learning during graduate school. I see SWE as my foundation and AI/ML as my specialization, allowing me to bridge production-level systems with advanced machine learning rather than treating them as separate paths.