FastAPI + Next.js ColPali Template
Full-stack Vision RAG template combining ColPali, Qdrant, FastAPI, and Next.js.
- RAG
- Vision
- ColPali
- Qdrant
- FastAPI
- Next.js
- Template
Hello, I'm Athos.
I'm an R&D Technical AI Lead specializing in Generative/Conversational AI, Machine Vision, RAG and Knowledge Graphs.
About me
I'm aR&D Technical AI Leadfocused on production AI -Generative/Conversational AI,RAG (text, vision, audio),Machine Vision,AI AgentsandKnowledge Graphs. I care about shipping practical, high‑impact systems that deliver measurable business value.
With 10+ years of experience across Financial Services, Energy, Legal and Academia, I currently lead AI R&D initiatives at CGI (Scotland & Northern Ireland) - from strategy and architecture through to enablement and delivery.
I build and write about open source. Recent work spans Vision RAG templates (ColPali + Qdrant), GraphRAG with Neo4j + Qdrant, and many more.
Skills & Expertise
🧠
Languages
- Python
- TypeScript
- Java
- JavaScript
- SQL
- GraphQL
- HTML/CSS
🤖
AI & Machine Learning
- PyTorch
- Transformers
- LlamaIndex
- ColPali
- ColQwen2.5
- Ultralytics (YOLO)
- BoxMOT
- Supervision
🧩
AI Platforms & Models
- OpenAI
- Anthropic
- Gemini
- Claude
- Ollama
- Grok
- Perplexity
🛠️
Web & Backend
- Next.js
- React
- Node.js
- Express
- FastAPI
- Redux
- REST APIs
🗄️
Databases & Data Stores
- Qdrant
- Pinecone
- Neo4j
- MongoDB
- MySQL
- PostgreSQL
- Redis
- Elasticsearch
- OpenSearch
- MinIO
☁️
Cloud, Infrastructure & DevOps
- Docker
- AWS
- Azure
- Vercel
- Grafana
- Prometheus
- Git
- GitLab
- CI/CD
🔐
Security & Authentication
- OAuth 2.0
- Active Directory
- 2FA
🎨
UI Libraries
- Tailwind CSS
- Shadcn/UI
- Framer Motion
Projects
Little Scripts
Curated utilities and experiments for AI workflows (RAG, FastAPI, multimodal, automation).
- Utilities
- Python
- RAG
- FastAPI
- Automation
NLP GraphRAG with Qdrant & Neo4j
NLP-centric GraphRAG experimentation using Qdrant vector DB and Neo4j.
- GraphRAG
- Qdrant
- Neo4j
- NLP
Video Annotation with RF-DETR & BLIP-2
Machine Vision pipeline for detection and captioning with RF-DETR and BLIP-2.
- Machine Vision
- RF-DETR
- BLIP-2
- Tracking
Grafana Alloy Example
Observability setup example using Grafana Alloy for metrics/logs pipelines.
- Observability
- Grafana
- DevOps
- Infra
Titanium (Next.js AI App Template)
Open-source Next.js AI application template with best practices baked in.
- Next.js
- AI
- Template
Latest from my blog
The Most Beautiful RAG: Starring ColPali, Qdrant, Minio and Friends
- RAG
- Vision
- ColPali
- Qdrant
- Minio
- FastAPI
- Next.js
ColQwen2.5 FastAPI Integration
- FastAPI
- Embeddings
- Qdrant
- ColQwen
Audio RAG with ColQwen2.5-Omni
- RAG
- Audio
- Multimodal
- Embeddings
- Python
The Most Beautiful RAG: Starring Colnomic, Qdrant, Minio and Friends
- RAG
- Vector Search
- ColBERT
- Late Interaction
- Python
My experience
CGI - Career Progression
Edinburgh, Scotland
R&D Technical AI Lead (Mar 2025 - Present)
• Spearheading CGI's AI initiatives across Scotland & Northern Ireland
• Leading strategic AI research and development projects across multiple industry sectors
• Driving real-world AI implementation that delivers measurable business impact
Principal Consultant (Mar 2024 - Mar 2025)
• Led enterprise AI initiatives and technical governance frameworks
• Developed and delivered comprehensive AI education programs focusing on generative and conversational AI
• Core contributor to the Taranta project and Titanium Project (Open Source Next.js AI application)
Lead Consultant (Mar 2022 - Mar 2024)
• Designed and implemented microservices architectures and REST APIs
• Led infrastructure improvements including performance monitoring and static code analysis
• Spearheaded technical governance initiatives and team development programs
Sr. Consultant (Mar 2021 - Mar 2022)
• Developed and supported enterprise microservices architecture
• Served as Scrum Master for development team of 8 engineers
• Implemented comprehensive automation and verification frameworks
Ambassador
Vercel - v0.dev
• I was selected as a v0 Ambassador for Vercel’s v0.dev, an AI-assisted builder
that turns natural-language prompts into production-ready UI and app code
(ideal for React/Next.js workflows).
• My role focuses on creating and sharing practical content, providing
early product feedback, and representing v0 positively across developer
communities. The ambassadorship reflects my ongoing public demos, articles,
and hands-on projects built with v0.
Ambassador
Qdrant - Open Source Vector Database
• Since April 2025, I’ve been part of the Qdrant Stars program, an ambassador
initiative recognizing active contributors in the Qdrant community.
• Qdrant is an open-source vector database designed for efficient similarity
search and semantic retrieval, powering AI and machine learning applications
Advisor
Lapis AI Consults - New York, United States
• Pro bono advisory role for businesses leveraging AI in product development
• Providing strategic insights for AI go-to-market strategies
• Supporting AI implementation across Finance, Energy, Legal & Academia sectors
Advisor
Brilliant Labs - Open Source AI Eyewear
• Providing pro bono advisory services for AI-powered open-source eyewear technology
• Contributing expertise in graph-based retrieval systems and AI integration
• Supporting innovative AI applications in wearable technology
Sr Software Developer
GE Power & Renewable Energy - Edinburgh, Scotland
• Developed solutions extending the PhasorPoint Platform for energy sector applications
• Led security initiatives to identify and resolve software vulnerabilities
• Applied Lean/Agile methodologies with focus on scalability and security
• Enhanced IEC-104 Protocol testing capabilities
Sr Software Developer
SolarWinds MSP - Edinburgh, Scotland
• Developed applications for Remote Monitoring & Management Platform
• Delivered microservices in continuous delivery environment
• Participated in technical interviews and production release oversight
• Contributed to performance enhancements through hackathon initiatives
Analyst Developer
Student Loans Company - Glasgow, Scotland
• Built web applications and APIs for government services
• Developed robust automated test suites for continuous deployment
• Applied agile methodologies and knowledge sharing practices
Software Developer
SQA – Scottish Qualifications Authority - Glasgow, Scotland
• Full SDLC experience across Windows and Linux environments
• Key developer of organization's first Continuous Integration application
• Delivered Java training workshops and technical knowledge transfer
• Built comprehensive release management and deployment processes
Application Developer - Research Computing
St. Andrews University, Scotland
• Developed LAMP stack applications for academic research
• Administered Linux servers and provided CKAN data management system support
• Supported open source data management initiatives
Consultant - NLP / Sentiment Analysis
xAmplifier - Northbrook, IL
• Led development of Customer Experience Management platform
• Applied quantitative and qualitative analytics for client insights
• Developed natural language processing solutions including Word Cloud implementation
• Created dynamic, data-driven reporting systems
Publications & Documents
Front-End Monitor and Control Web Application for Large Telescope Infrastructures: A Comparative Analysis
The paper compares the Grafana and Taranta web applications as front-end GUIs for monitoring and controlling large radio telescopes using the TANGO Controls framework, evaluating their architectures, usability, and performance to guide stakeholders’ future choices.
Improving Performance of Taranta: Analysis of Memory Requests and Implementation of the Solution
The paper analyzes performance issues in Taranta’s increasingly complex, no-code Tango dashboards, particularly excessive client-side memory use and proposes a new React-based architecture, backed by tests and implementation plans, to make large, data-heavy interfaces more efficient.
Spatially-Grounded Document Retrieval via Patch-to-Region Relevance Propagation
The paper proposes a training-free hybrid retrieval method that uses ColPali’s patch-level vision-language similarities to filter and rank OCR-extracted text regions, enabling more precise, region-level context for RAG, and introduces Snappy as an open-source implementation of this approach.
Contact me
Please contact me directly at athrael.soju@gmail.com or through this form.