I'm a Computer Science and Electrical Engineering student at Grambling State University (GPA 3.88, graduating May 2028). I like the layer where software meets the metal — packet pipelines that have to close timing, CUDA kernels that have to beat the CPU, agents that have to price their own uncertainty. I build it, then I try to break it and write down what actually happened. The problems_encountered.md in my repos is usually the most interesting file.

🌐 Portfolio | 💼 LinkedIn | 📧 iminaboroberts@gmail.com

May 2026 – Aug 2026

• Built a recursive self-improving loop for Xcode coding agents that consolidates and replays their own history.
• Designed a heartbeat that continuously updates and merges stored context to steer generation.
• Built an evaluation harness benchmarking agent performance — measured a 69% gain at lower cost.

April 2026

• Added static IP support to Google TestRun, which previously handled only DHCP assignment.
• Investigated and resolved test-execution inconsistencies by debugging across distributed workflows.
• Collaborated with maintainers through issue discussions and code reviews to align fixes with project architecture.

Aug 2025 – Dec 2025

• Scaled backend throughput 30% by running PyTorch/HuggingFace inference on EKS with gRPC and S3-backed services.
• Automated retraining on AWS SageMaker with datasets in S3, ensuring continuous model updates.
• Reduced downtime by implementing monitoring with Grafana.

Sep 2025 – Present

• Investigating temperature and surfactant effects on nanoparticle size measurement via Dynamic Light Scattering.
• Statistical validation and modeling in Python (NumPy/Pandas), improving nanoparticle characterization accuracy for laser-ablation research.

Mar 2025 – Apr 2025

• Applied game theory and decision theory to price auction bids under strategic uncertainty.
• Ranked strategies by probability-weighted expected value on z-scored surfaces in pandas.
• Placed in the top 10% of 12,000+ teams.

Jan 2025 – Apr 2025

• Built a chest-worn heartbeat monitor in C++ on Arduino, streaming live waveforms to a display.
• Detected arrhythmias via real-time signal analysis using I2C and oscilloscope diagnostics.

Sep 2023 – May 2024

• Earned a spot in a competitive program with a 21% acceptance rate.
• Enhanced robot mobility 75% developing in Java on Ubuntu with Raspberry Pi.
• Mentored a Nigerian FIRST Robotics team on low-level concepts, Android Studio, the FTC SDK, and Onshape CAD.

SystemVerilog

• A SystemVerilog pipeline parsing Ethernet / IPv4 / UDP to drop, forward, timestamp, or rewrite.
• Pipelined a 4-stage datapath on 64-bit AXI-Stream at one word per cycle, loading flow tables over AXI-Lite.
• Verified with randomized cocotb testbenches, targeting 250 MHz while trading pipeline depth against latency and LUTs.

OCaml & Unix sockets

• An OCaml tile server where image tiles bid for compute, so high-detail boundary regions render first.
• Parallelized across N Domain workers pulling from a bid-sorted queue via a lock-free atomic cursor.
• Cut render time 2.44× by letting drained workers steal across markets — utilization went from 35% to 88%.

Python & C++

• Built a WGAN-GP and a diffusion model in PyTorch on FI-2010 limit order books.
• Raised generated book validity from 0% to 100% with a cumulative-softplus gap generator — validity by construction, not by penalty.
• Evaluated on spread, depth, and order imbalance against real market data.

Python & C++

• Shipped a heuristic agent to a live Kaggle ladder of roughly 5,000 competing teams.
• Built Perfect Information Monte Carlo search over the competition's C++ forward model, averaging over sampled hands.
• Modeled hidden opponent hands via determinization sampling, weighting guesses by observed play patterns.

C++ & Python

• A C++ order book engine publishing trades to Kafka with asynchronous librdkafka callbacks.
• Handled concurrency with one thread per instrument bridged into an asyncio event loop.
• Streamed real-time market data over gRPC and WebSocket with backpressure on a 100 ms timeout.

C++ & CUDA

• Shi–Tomasi corner detection on CUDA using GpuMat, cudaimgproc, and cudaarithm.
• 60 FPS on an RTX 3050 — 4× faster than the CPU baseline via optimized memory transfers.

Python & Java

• A multimodal assistant combining screen vision, LLM parsing, and voice commands to automate app workflows.
• Semantic search across the device file system using embeddings to contextualize prompts.
• Streamed audio output before model completion for sub-150 ms time-to-first-response. Extending to Android via Appium.

Python

• A Playwright agent deriving read and write MCP tools from a URL via the accessibility tree.
• Parallelized network analyzer generating async Python tools, filtering 70% of noise.
• Self-healing auth for CSRF and cookies, eliminating manual agent re-login.

C++ & Python

• Bilateral knee sensors in C++ on Arduino Nano BLE streaming IMU data over Bluetooth.
• Trained an LSTM risk model on movement time series to flag injury risk against baseline patterns.
• Computed 5 movement metrics from raw IMU samples at 200 Hz on device, streamed to a Vercel dashboard over WebSockets.

JavaScript & Python

• A Chrome extension injecting AI nudges via Node.js, Tailwind, and the OpenAI/Claude APIs.
• Supabase memory with embedding search, recalling past interactions to personalize guidance.
• Sub-200 ms hint latency and 65% distraction reduction.

• FIRST Tech Robotics Challenge — National Winners · Inspire Award
• National Informatics Olympiad — Finalist
• Putnam competitor
• Microsoft Imagine Cup — Pre-Semifinalist
• Notion Hackathon — 3rd place · GDSC Coding Competition — 2nd place
• Nexus Louisiana DevDays ClimateTech — Finalist · BeSmart Hackathon — Finalist

I believe in lifting others as I climb. Technology is empty without community, and these are the spaces where I pour my energy outside of building software:

• ColorStack @ Grambling State — Academic Chair (Jan 2025 – Present): Supporting Black & Latinx computer science students throughout their college careers; organizing workshops and helping members work through hard technical problems.
• NSBE @ Grambling State — Community Service Chair (Feb 2025 – Present): Coordinating STEM outreach events and mobilizing members to mentor local students.
• Grambling Robotics Club — Vice President (Aug 2025 – Present): Leading weekly design-and-prototyping sessions; training peers in Arduino IDE, ROS2, Fusion 360, Onshape CAD, and TinkerCAD.
• Competitive Programming Club — Head: Built CF:ICPC, a training platform for our 8-person ICPC team — ~220 Codeforces problems in a prerequisite-gated skill tree, per-member progress tracking, and a 10,700-problem similarity graph. A "Codeforces 220" built for ICPC's mathematical breadth rather than interview pattern-matching.
• Programs & fellowships: Bloomberg Accelerator Fellow · Intel Student Ambassador · NVIDIA Summer Bridge · eBay Pathways · LS-LAMP Research Intern · Brilliant Black Minds (Karat) · Earl Lester Honors College · ACM · Lantern Network · General Assembly Immersive

• Teaching models to play games for fun: Super Mario Bros, Sonic 2, and the Google Dino game — all reinforcement learning, all slightly cursed.
• Writing things down: every serious repo of mine ships a journal.md and a problems_encountered.md. The failures are the useful part, and I'd rather publish them than pretend the first approach worked.
• Robotics for its own sake — FIRST alum, and I still like putting sensors on things that don't strictly need sensors.