AI engineer
Mohamed Sameh
محمد سامح
I build generative vision pipelines, document AI and tool-using agents, and the interfaces that let people check what the model did.
- A+IntelliRoom, graduation project
- 1stSTP MACHATHON 7.0, document AI
IntelliRoom shop-the-look A catalogue bed placed into the photo
Editable table 18 rows × 6 columns, 108 cells read
| 410 State Textbook | 429 State Funded Special Revenue | 486 Central Office | 487 Leading & Learning | 489 Athletic Activity | 490 BHS Campus Activity |
|---|---|---|---|---|---|
| $ | $ | $ 79,450 | $ 2,385 | $ 383,911 | $ 61,244 |
| 172,014 | 228,474 | 1,366 | |||
| 172,014 | 228,474 | 79,450 | 2,385 | 385,277 | 61,244 |
| 202,034 | 393 | 28,652 | |||
| 4,944 | |||||
| 10,318 | 1,748 | ||||
| 19,222 | |||||
| 111 | 489,965 | 2,814 | |||
| 211,173 | |||||
| 76,814 | |||||
| 202,034 | 221,884 | 76,925 | 1,748 | 489,965 | 55,632 |
| (30,020) | 6,590 | 2,525 | 637 | (104,688) | 5,612 |
| 121,045 | 1,000 | 262,149 | 41,858 | ||
| $ 91,025 | $ 6,590 | $ 2,525 | $ 1,637 | $ 157,461 | $ 47,470 |
CSVExcelHTML
Agent P-DF Detect the table, recover rows and columns, read every cell
Selected work
Eight projects from 2025 and 2026, in generative vision, document AI, agents and applied computer vision. Each case study opens with the result.
-
Graduation project, graded A+
IntelliRoom
Edits room photos, places catalogue products into them and builds 3D models from product photos. I built every AI workflow and the catalogue.
- 72 node graph behind nine switches
- 158 catalogue products generated
-
Document AI, 1st place at STP MACHATHON 7.0
Agent P-DF
Finds tables in scanned pages, rebuilds their rows and columns, reads the text and opens the result in an editor before export.
- 1st place, STP MACHATHON 7.0
- 5,353 wrong training labels removed
-
Agentic security, exploratory
MalWhere
A local language model triages Windows executables, calling nine analysis tools one at a time and streaming every step.
- 9 tools, 7 local and 2 threat feeds
- 0 samples executed
-
Concurrent agents, research
M.A.C.E.
A Minecraft agent where a supervisor checks for danger every three seconds and can abort the skill the agent is running.
- 0.52 s median code generation
- 3 s between danger checks
-
Learning workspace, my own product
MoaLLM Board
You and an AI agent study side by side on one canvas: it explains, examines you and marks the paper, and searches your notes with an embedding model I distilled for it.
- 0.3 ms per query, own embedding model
- 60 agent tools over MCP
-
Game AI, personal project
KotchAIna
A Basra card game engine whose AI estimates the opponent's hidden hand, solves the endgame exactly and explains each move.
- 302 passing tests
- Exact endgame solver
-
Android, desktop and Firefox, open source
Zakkir
Prayer times and adhkar in Arabic, with reminders that keep working offline once the times are cached.
- 138 adhkar in 8 categories
- 28 days of reminders scheduled offline
-
Ain Shams University, Faculty of Engineering, 2025
PhotoValidator
Sorts image batches into valid, invalid and manual review, built during a training task with the faculty's IT team.
- 5 checks per image
- 3 outcome folders
Ain Shams, class of 2026.
I studied Computer and Systems Engineering at Ain Shams University in Cairo. My work centres on generative AI systems, multimodal models and agent workflows, built end to end, from fine-tuning and data repair to the interface someone uses.
- Degree
- B.Sc. Computer and Systems EngineeringFaculty of Engineering, Ain Shams University, 2021 to 2026
- Graduation project
- IntelliRoom, graded A+Generative interior design platform
- Award
- 1st place, STP MACHATHON 7.0Document AI table extraction, 2026
- Shared task
- AISA-ArabicFC, ArabicNLP 2026Arabic function calling and agent reasoning, ongoing
- Certification
- Data Analysis, National Telecommunication InstituteTop leaderboard ranking, 2024