FH-2027 · PRELIMINARY DATASHEET · REV 2.1

Faris Hittiny

Building computer systems from register-transfer level to production AI.

  • AI/ML ENGINEERING INTERN · ARAMCO AMERICAS
  • CUSTOM RISC-V ACCELERATOR · 15.1× FEWER CYCLES (SIM)
  • COMPUTER ENGINEERING '27 · TEXAS A&M UNIVERSITY
0x000 · IDENT

About

I'm a computer engineering senior at Texas A&M University, graduating May 2027. I work at the boundary where hardware meets software: the layer where a design decision shows up as real cycles, real watts, real latency.

That range is deliberate. I've written Verilog for a custom accelerator and benchmarked it cycle-by-cycle in simulation; I've also shipped AI features inside enterprise software where the constraint isn't the model, it's the system around it. The common thread is wanting to know why the number is the number: measuring before claiming, and being honest about what a benchmark does and doesn't say. The same instinct runs the rest of my life. Ask anyone who's seen how I organize a pickup soccer bracket.

  • LOCATION COLLEGE STATION, TX
  • DEGREE B.S. COMPUTER ENGINEERING, MAY 2027
  • FOCUS HW/SW CO-DESIGN · APPLIED AI/ML
  • STACK C/C++ · PYTHON · VERILOG · C# · TS
Portrait of Faris Hittiny
FIG 1.1 · SUBJECT, BUSINESS FORMAL
0x004 · STATUS

Experience

AI/ML Engineering Intern

Aramco Americas

SUMMER 2026 · HOUSTON, TX

Two tracks, both production-facing. The main one was AI invoice extraction on Azure OpenAI: pulling line items out of AT&T, Verizon, and T-Mobile bills for 500+ corporate lines, with anonymized datasets and validation pipelines that measured 99.8% extraction accuracy against 8+ years of production billing data. Around it, I shipped feature work on the internal telecom expense platform (ASP.NET, C#, SQL Server), including a bulk employee-status refresh I cut from 6.5 to 3.3 minutes by tuning API concurrency.

On the second, I built agents on Salesforce Agentforce: designing agent actions, grounding them in org data, and testing the failure modes that separate a demo from something a business unit can actually rely on.

The NextGen Leaders program sent me home with an Excellence in Teamwork award. The certificate is larger than my head.

  • AZURE OPENAI
  • ASP.NET / C#
  • SQL SERVER
  • SALESFORCE AGENTFORCE
Faris and two fellow interns in hard hats and safety glasses at an industrial field site
FIG 2.1 · SITE VISIT, PPE PER SPEC
0x008 · RESULT

Research & Projects

RISC-V Hardware Accelerator

A custom AXI-Lite popcount accelerator, integrated into a LiteX/VexRiscv SoC, driven by a Zephyr driver, and benchmarked cycle-accurately under Verilator.

  • VERILOG
  • LITEX / VEXRISCV
  • AXI-LITE
  • ZEPHYR RTOS
  • VERILATOR
15.1× FEWER CYCLES VS SW BASELINE, 4 KB
~1 B/cyc CORE THROUGHPUT AT DESIGN RATE
0xA0000000 MEMORY-MAPPED, 8 KB WINDOW

MEASURED IN CYCLE-ACCURATE LITEX/VERILATOR RTL SIMULATION, NOT SILICON. BASELINE IS THE SHIPPED BYTE-WISE SW LOOP, NOT AN OPTIMIZED SWAR KERNEL.

github.com/FarisHittiny/riscv-accel →

Catalog RAG

Eval-driven retrieval over the Texas A&M course catalog: BM25, embeddings, and a prerequisite graph behind a query router, every change measured against a hand-verified gold set before it was kept.

  • PYTHON
  • BM25
  • BGE EMBEDDINGS
  • NETWORKX
  • FASTAPI
  • CLOUD RUN
0.61 → 0.90 RECALL@5 VS BM25 BASELINE
0.51 → 0.79 ANSWER CORRECTNESS (LLM JUDGE, 90% HUMAN AGREEMENT)
129 QUESTIONS · HAND-VERIFIED · HELD-OUT VALIDATED

MEASURED ON 148 ECEN/CSCE COURSES. DENSE ROUTE DESIGNED POST-HOC, THEN VALIDATED ON A 14-QUESTION HELD-OUT SET. NUMBERS WILL DROP AS THE CORPUS GROWS.

Cattle Eartag Detection Research

Computer-vision research for livestock monitoring: detecting and reading cattle eartags in field imagery as groundwork for earlier disease detection. Built detection pipelines with YOLOv8 and GroundedSAM, processed and labeled large-scale image datasets, and automated preprocessing that had been manual. Presented at the 2025 Undergraduate Research Expo.

  • YOLOV8
  • GROUNDEDSAM
  • PYTORCH
  • OPENCV
Presenting eartag-detection research at the 2025 Undergraduate Research Expo
FIG 3.1 · 2025 UNDERGRADUATE RESEARCH EXPO
0x00C · CTRL

Leadership

Membership Chair

Arab Student Association, Texas A&M

2026–PRESENT

I run membership for ASA: recruitment, onboarding, and making sure the people who show up once come back. It's operations work with a human core: tabling season, first-event welcomes, and keeping a growing roster feeling like a community rather than a mailing list.

ASA members around a table mid-card-game at game night
FIG 4.1 · ASA GAME NIGHT, RETENTION STRATEGY IN PRACTICE
0x010 · CONFIG

Off the Clock

Recommended operating conditions. Exceeding these ratings is common and encouraged.

Interests, presented as a datasheet operating-conditions table
PARAMETERMODENOTES
Pickleball PLAYER Always up for a game. Dinks are a legitimate strategy and I will not be argued out of this.
Soccer PLAYER / ORGANIZER I play, and I'm usually the one wrangling the group chat into an actual match.
Live sports SPECTATOR If there's a game on somewhere nearby, odds are decent I'm in the stands. Career highlights: Jordan's first-ever World Cup in Dallas (FIG 5.2) and meeting Roberto Carlos next to the actual trophy (FIG 5.1).
Faris giving a thumbs-up next to Roberto Carlos and the FIFA World Cup trophy
FIG 5.1 · SUBJECT (LEFT), ROBERTO CARLOS (RIGHT), ONE (1) WORLD CUP TROPHY
Faris and friends in Jordan kits holding a We Are Jordan flag in a packed stadium
FIG 5.2 · JORDAN'S FIRST WORLD CUP, DALLAS
Aramco interns in national-team soccer jerseys at flag day, Faris in a Jordan kit
FIG 5.3 · FLAG DAY, ARAMCO HOUSTON (SUBJECT IN JORDAN KIT)
0x014 · IO

Contact

Internships, research, projects, or a pickup game: my inbox is open.