
Wahiduzzaman Suva
Applied AI Engineer
I build production AI systems such as agent runtimes, document intelligence pipelines, and retrieval infrastructure. Alongside engineering, I conduct research in brain-computer interfaces and quantum machine learning.
Journey

Joined Odin, which has supported financial data processing for the US commercial real-estate market since 2013: loan origination underwriting and acquisition analysis for investment banks, REITs, and private owners. I build the tooling behind that reporting: a model plans a task, calls tools that read from and write directly into the analysts' live Excel workbooks, verifies its output against the source documents, and re-runs the fields it got wrong. Underneath sits a per-page extraction cascade (pdfplumber, Camelot lattice and stream, Tesseract OCR) covering PDF, Word, and scanned files, scored to keep the best result per page. Everything runs on client hardware, because bank documents can't leave the machine.

Opened Prodigy AI with a group of friends from AIUB, a non-profit research organization exploring neuroscience and AI: brain-computer interfaces, cognitive computing, and neuroimaging analysis. Open-source projects for collaboration, private research for focused studies.

Started Byteverse, an IT solutions company working across AI development, software, cloud, fintech, and data analytics. By early 2025 we had delivered 30+ projects for clients across the US, UK, Australia, and India, growing from two founders to a full team. I work on Chatsby, our customer-support AI product: the agent runtime, retrieval over Qdrant, realtime chat, and the deployment stack (FastAPI, Redis, Nginx, GitHub Actions). Along the way I did contract backend work for client products: a WhatsApp marketing agent, an outbound calling pipeline, comment and DM automation, and a course platform on AdonisJS. Also proud of aipedia.blog and sspark.

BSc in Computer Science and Engineering at AIUB, finishing with a 3.97 CGPA. Started the MSc there in 2025.

Left for Saint Petersburg State University intending a software engineering degree; earned TORFL B1 Russian along the way, then discontinued and came home, putting the Russian to work as a Russian-Bangla interpreter at the Rooppur Nuclear Power Plant.
Writing
News
- Jan 2026Paper on equivariant quantum clustering published in Physica Scripta. linkView post on LinkedIn
- 2026VC-DATNet, a dual-attention transformer for cross-dataset inner-speech decoding, is under review.
- 2025Started MSc in Computer Science at AIUB.
- Nov 2025Book chapter on federated learning in healthcare published (Springer SCI vol. 1216). link
- Oct 2025Recognized as Rising Star Performer 2025 at ODIN Outsourcing.

- Sep 2025Chatsby, the customer-support AI product we build at Byteverse, launched. View post on LinkedIn
- May 2025Presented our EEG transformer research at AIUB. link

- Mar 2025Quantum graph neural network paper published in IEEE Transactions on Artificial Intelligence. linkView post on LinkedIn
- Feb 2025Joined Odin.
- 2024Opened Prodigy AI, a non-profit research group working on brain-computer interfaces. link
- 2024First paper published, in Metaversalize. link
- 2024Organized the national round of TechnoXian Bangladesh, a robotics competition.

- Apr 2024Presented Coder's Space, our developer dashboard, at AIUB CS Fest. link





- Jun 2023Joined Byteverse.
Projects
Up-to-date code lives on my GitHub.

Chatsby is Byteverse's customer-support AI platform. I work on the agent layer, retrieval over Qdrant, and the realtime and deployment stack (FastAPI, Redis, Nginx). See the launch post.

Outbound Calling Assistant is a calling pipeline with lead scoring, live transfer to a human, voicemail detection with retry, and CRM writeback, built on the agent backend of a WhatsApp marketing product.

CEO Society is a course platform whose backend I built: courses, cohorts, roles, and an in-product assistant that answers from course material.

CommentZap automates comments and DMs on the Meta Graph and Messenger APIs: webhooks, rule-driven responses, token handling, rate-limit backoff.

Acquisition Engine is Byteverse's lead acquisition engine: scraping with validation, running on Docker and Redis.

Financial Asset Suite is a suite of micro-services for financial reporting at Odin, working directly in Excel workbooks on client infrastructure: Cashflow AI (cashflow models), Rent Roll Analyst (rent roll standardization), BossPDF (document workflows), Pexl (table and field extraction from appraisals, bank statements, and utility bills with pdfplumber, Camelot, and Tesseract), a vision model for QC, and Market and Submarket analysis.
Publications
Md Abrar Jahin, Md. Akmol Masud, Md Wahiduzzaman Suva, M. F. Mridha, Nilanjan Dey
Md Wahiduzzaman Suva, Md. Imtiaj Alam Sajin, Esm E Moula Chowdhury Abha, Md Tauhidul Islam, M. F. Mridha, Md. Abdul Hamid, Md. Jakir Hossen
Esm E Moula Chowdhury Abha, Md. Imtiaj Alam Sajin, Md Wahiduzzaman Suva, Shanjida Akter, Mushfiqur Rahman Abir, Md. Asgor Hossain Reaj, Jahir Ibna Rafiq
Md Wahiduzzaman Suva, Md. Imtiaj Alam Sajin, Esm E Moula Chowdhury Abha, Mushfiqur Rahman Abir, Asif Zaman
Also on Google Scholar
Academic Projects

VC-DATNet: Inner-Speech Decoding. Variance-coupled dual-attention transformer for cross-dataset inner-speech decoding from EEG. Pairs a shared multi-scale temporal-spatial front-end with parallel mean and log-variance encoder streams of dual-attention transformer layers. Every number in the manuscript is reproducible from raw OpenNeuro data (ds003626, ds004196, ds006033).
CB-SAFE: Post-Quantum Secure Aggregation for Federated Learning. Code-based (HQC) post-quantum secure-aggregation framework for federated learning. KEM-agnostic cluster-based secure aggregation with Bonawitz-style double masking, Shamir dropout recovery, and robust aggregation rules (trimmed mean, median, Multi-Krum) against poisoning attacks, evaluated on non-IID CIFAR-10 with 30 clients.

Koala Tracklog Analysis. Analysis of koala movement patterns from GPS tracklog data. Preprocessed GPS tracklogs with Pandas and NumPy, visualized with Matplotlib, Seaborn, and Plotly, and applied Scikit-learn models to identify movement trends and behavioral patterns.


Equivariant Quantum Clustering (EQC). p4m equivariant 8-qubit quantum clustering framework behind the Physica Scripta paper. Reference implementation for A symmetry-aware equivariant quantum clustering approach for enhanced unsupervised learning (Physica Scripta 101(2) 026007, 2026, DOI 10.1088/1402-4896/ae2dd7). On MNIST it reaches silhouette 0.557 and 76.3% accuracy; on Quark-Gluon, 0.512 and 74.8% against 0.183 and 58.4% for K-Means.



Face Recognition with FaceNet and MTCNN. Face recognition using FaceNet embeddings and MTCNN detection. Maps face images into a 128-dimensional embedding space where distances correspond to face similarity, with MTCNN for detection.

Sentiment Analysis on Conti Ransomware Leaks. Sentiment analysis of the leaked Conti ransomware chats to uncover trends and emotional tone. Cleaned and structured the leaked data with Pandas, classified message sentiment with VADER, explored relationships between actors with NetworkX, and visualized findings with Matplotlib and Seaborn.

Space Rover Application. Control and telemetry application for the AIUB Robotics Crew Mars rover. Real-time sensor data integration and processing with onboard image analysis for rover operations.




dataviz: Statistical Visualization in R. R workflows behind the Metaversalize paper on advanced visualization for statistical data analysis. Demonstrates the paper's methods on the UCI HCV dataset (Metaversalize 1(4), pp 225-239, 2024, DOI 10.22105/metaverse.v1i1.36).



LIFT Ride Sharing Database. Relational schema for a mixed-fleet ride sharing service. A single vehicle table with an owner_type discriminator covers company and driver owned fleets, and payouts settle in batches. Tested on SQLite.
Skills
Activities
Research group working on brain-computer interfaces. EEG decoding work: Brain-Controller and VC-DATNet.
Facilitated effective communication with government officials and the HR department. Prepared documentation for work permits, legalization, and visas for foreign employees. Attended multi-party meetings as a Russian-Bangla interpreter.


Designed the rover's science module (2023-24) and work on soil analysis, biochemical testing, and scientific payload development for the University Rover Challenge, alongside sensor data processing in Python and OpenCV.

Misc
- I recorded a few tutorials during university: a Java interfaces lecture, and OpenGL/GLUT guides on texturing shapes and loading images. More on YouTube.
- I used to make games: a first-person adventure in Unity about finding keys to complete missions, a canyon mystery game, and Galactic Exodus: Homeward Bound, a space game written from scratch in C++ and OpenGL (source).
- For two years I organized workshops and exhibitions with the AIUB Arts Club: 15+ events like the Intra-AIUB Art Exhibition, running logistics, sponsorship, and promotion.
- Opened Runique, a shoe brand, with Imtiaj Sajin. We sold roughly 280 pairs of shoes until our page got banned XD
Contact
© 2026 Wahiduzzaman Suva · wshuv-o.github.io





