Hi, I'm
Shahriar Soudeep
Building production AI systems — from RAG pipelines to published research.
About
A little about me
AI Developer at Softvence Agency, building production-grade LLM services, RAG pipelines, and multi-agent systems. Concurrently a Research Assistant at the AMIR Lab under Prof. Dr. M. Firoz Mridha at AIUB, contributing to peer-reviewed research spanning graph neural networks, transformer architectures, and explainable AI. Graduated with a CGPA of 3.79/4.00 in Computer Science & Engineering.
American International University–Bangladesh (AIUB)
B.Sc. in Computer Science & Engineering · 2020 – 2024
CGPA 3.79 / 4.00
Experience
Where I've worked
Production AI systems and applied research, side by side.
AI Developer / AI Engineer
Softvence Agency
- Led design and deployment of production-ready FastAPI backends powering LLM-integrated services, computer vision pipelines, and OCR systems
- Built RAG-based systems with persistent memory using LangChain, FAISS, Redis; orchestrated multi-agent workflows via CrewAI and LangGraph
- Designed and deployed N8N-based automation workflows integrating webhook triggers, third-party APIs, and CRM systems
Research Assistant
AMIR Lab
- Contributed to 6 peer-reviewed publications spanning GNNs, transformer architectures, VLMs, and explainable AI (IJCAI, IEEE DSAA, IEA/AIE, Q1 journals)
- Currently researching post-occlusion material-point reacquisition in 3D point tracking
- Leading development of Bangla-Bayanno VQA, a large-scale Bengali visual question answering dataset
Computer Vision Engineer (Contract)
Tiva Solution
- Delivered production YOLOv8 detection system at 47 FPS; reduced false-positive rate by 22% via post-processing tuning
- Implemented image processing pipelines using OpenCV, MediaPipe, Haar Cascade classifiers
- Designed FastAPI endpoints serving inference results as REST responses
Projects
Things I've built
Production systems shipped for clients and internal platforms.
Orchestrated 14 AI Interaction Points using Claude API for skill assessment, real-time feedback, and personalized learning pathways via an auto-sequencing state machine.
Reduced API calls by 60% via smart caching & fallback strategy.
Automated Client Acquisition & Lead Pipeline
Fully automated N8N workflow orchestrating multi-step LinkedIn outreach — generated 491 qualified leads with zero manual intervention.
Trigger-based logic connecting Dripify, Calendly, and Google Sheets via webhooks, with a real-time CRM-style dashboard.
Research
Publications
Peer-reviewed work spanning GNNs, transformers, VLMs, and explainable AI.
Vision Transformers for End-to-End Quark-Gluon Jet Classification from Calorimeter Images
A*IJCAI 2025
View PaperDyCAF-Net: Dynamic Class-Aware Fusion Network
Core-AIEEE DSAA 2025
SOTA precision/mAP@50-95 across 13 benchmarks at ~11.1M params
View PaperPhysics-Informed Graph Neural Networks for Transverse Momentum Estimation in CMS Trigger Systems
Q1Computer Physics Communications, Elsevier · 2026
SOTA MAE of 0.8525 with ≥55% fewer params than baselines
View PaperEnhancing Road Traffic Flow in Sustainable Cities through Transformer Models
Q1Sustainable Cities and Society, Elsevier · 2023
View PaperCAGN-GAT Fusion: A Hybrid Contrastive Attentive Graph Neural Network for Network Intrusion Detection
Core-BIEA/AIE 2025
View PaperSkills
Tools & technologies
The stack I use to design, build, and ship AI systems end-to-end.
AI / GenAI
RAG & Vector
Backend / API
ML / CV
Automation
Cloud & DevOps
Databases
External APIs
Languages
Contact
Let's work together
Open to full-time roles, research collaborations, and AI automation engineering work.