Souvik Basu

Solutions Architect (Applied AI) | LLM Integrations & Evals | Python | AWS/GCP

AWS Certified AI Practitioner

15+ years delivering cloud-native backend and data systems, recently focused on Applied AI, LLM APIs, and production automation

View Projects Get In Touch
Souvik Basu

About Me

Solutions Architect with 15+ years delivering cloud-native backend and data systems, recently focused on Applied AI (LLM APIs, HuggingFace/Transformers, LangChain) and production automation. Hands-on Python, AWS (primary) with GCP exposure, and strong stakeholder communication for fast-moving teams.

Work Authorization: Skilled Worker Visa (UK) — ILR eligible May 2027 (sponsorship transferable)

Skills & Expertise

Python
Node.js
PHP
JavaScript
Java
AI Integration
LLM Integration
OpenAI APIs
OpenAI Vision
Claude
Groq
Multiple LLMs
Workflow Automation
HuggingFace
BERT Models
Fine-tuning
SDXL
ControlNet
LoRA
ComfyUI
AWS
Google Cloud
Docker
Kubernetes
CI/CD
AWS CodeDeploy
ETL Pipelines
Data Scraping
Selenium
BeautifulSoup
MySQL
MongoDB
Data Processing
React
Vue.js
Tailwind CSS
REST APIs
Microservices
Runpod
Windsurf
Git
Agile
QA Automation

Experience

Twyzle – Senior Software Developer

December 2021 – Present (4 years) | Basingstoke, England

Senior Full-Stack Engineer | AI & Systems Architect

AI & NLP Innovation

  • Built hybrid job classification pipeline using HuggingFace Transformers, fine-tuned BERT models, and GPT-based classifiers
  • Implemented zero-shot classification for niche job titles with precision-focused rules and embeddings
  • Developed feedback/logging loop to auto-flag misclassifications and iteratively improve model accuracy

Data & ETL Pipelines

  • Designed scalable scraping pipelines (Selenium, BeautifulSoup) across multiple job boards
  • Created ETL flows to clean, transform, and post data into MySQL/MongoDB with validation and monitoring layers

Generative AI Projects

  • Fine-tuned AI models (SDXL, ControlNet) using LoRA and cinematic prompts for custom image generation
  • Built auto-captioning pipelines with BLIP2 and image conditioning workflows using ComfyUI

Fresh Gravity – Manager API Management & Integrations

July 2019 – January 2022 (2 years 7 months) | Pune Area, India

  • Responsible for defining overall QA Automation strategy (Performance, UI)
  • Accountable for client communications, multiple QA automation projects, mentoring and providing leadership to teams
  • Handling projects with multiple technologies like AWS, Java, Python, Selenium

Independent Consultant

September 2012 – June 2019 (6 years 10 months) | Kolkata Area, India

Independent Consultant and web solution architect

IBM Global Services – Senior System Developer

January 2012 – October 2012 (10 months)

  • Provide Support and maintenance (Include new developments) of Perl/Java applications and take full ownership
  • Create new tools for monitoring and automate manual processes using Perl
  • Provide ideas on how to improve application performance

HSBC – Senior Software Developer

August 2006 – August 2009 (3 years 1 month)

  • Developing web applications that meet both business requirements and standards
  • Maintenance and Development of Web Development Environment
  • Design and implementation of JavaScript Framework and Perl Framework
  • Technical consultant on all matters impacting applications

creativeskills – Software Engineer

January 2005 – July 2006 (1 year 7 months)

  • Involved in development of client-side JavaScript
  • PHP coding of the Client as well as Admin side
  • Integration of Components

Featured Projects

AI-Powered Conversational Website Builder

A full-stack web application that revolutionizes website creation through natural language conversation. Built with Flask, Python, and modern frontend technologies.

Key Features:

  • Conversational Interface: Natural language website generation
  • Template System: Professional templates (SaaS, Restaurant, Portfolio, E-commerce)
  • Real-time Updates: Interactive modification system
  • Smart Image Handling: Uploads, URLs, and AI-generated visuals
  • Reference Integration: Document/website context-aware design

Technical Stack:

  • Backend: Flask with LLM integration (OpenAI, Google Generative AI, Ollama)
  • Frontend: Modern JavaScript with Tailwind CSS
  • AI Integration: Multiple LLM providers with fallback mechanisms
  • Image Processing: FLUX.1-dev, Stable Diffusion, Unsplash API

Document Q&A System

Intelligent Document Question-Answering with Multi-Modal AI. Chat with your documents (PDF, DOCX, Images), leverage powerful RAG, and toggle between local (Ollama) and cloud (Gemini) AI models.

Key Features:

  • Multi-format document support (PDF, DOCX, Images, Web content)
  • Flexible AI providers (Local Ollama + Cloud Gemini)
  • Hybrid search (semantic + keyword matching)
  • Image understanding with OCR and Vision AI
  • Multi-project management with persistent conversations
  • Source attribution and export capabilities (JSON, CSV, PDF)

Technical Stack:

  • Framework: LangChain + Streamlit UI
  • Local Models: Ollama (Qwen 2.5, LLaVA)
  • Cloud: Google Gemini 2.5 Flash
  • Vector DB: ChromaDB with hybrid search
  • OCR & Vision: Tesseract + EasyOCR
  • Database: SQLite with SQLAlchemy ORM

AI Job Classification

Hybrid classification engine using HuggingFace Transformers, fine-tuned BERT models, and GPT-based classifiers with feedback learning loop for accurate job title and category classification.

Key Features:

  • Hybrid classification engine combining multiple approaches
  • Fine-tuned BERT models for semantic understanding
  • GPT-based classifiers for complex job titles
  • Zero-shot classification for niche job titles
  • Feedback learning loop for continuous improvement
  • Auto-flagging of misclassifications
  • Precision-focused rules and embeddings

Technical Stack:

  • Framework: HuggingFace Transformers
  • Models: Fine-tuned BERT, GPT-based classifiers
  • Embeddings: Semantic embeddings for similarity matching
  • Database: MySQL/MongoDB for classification history
  • Feedback System: Auto-learning from misclassifications

ETL Scraping Pipelines

High-scale scraping engines using Selenium, BeautifulSoup across multiple job boards with ETL flows to MySQL/MongoDB. Data is transformed into golden datasets and distributed to subscribers across various domains (jobs, cars, ads, etc.).

Key Features:

  • Multi-source scraping (Selenium, BeautifulSoup)
  • Distributed scraping across multiple job boards
  • Data extraction, cleaning, and validation
  • ETL transformation pipelines
  • Golden data lake for normalized datasets
  • Subscriber-based data distribution system
  • Domain-specific data feeds (jobs, cars, ads, etc.)
  • Real-time and batch processing modes
  • Error handling and retry mechanisms
  • Data quality monitoring and logging

Technical Stack:

  • Web Scraping: Selenium, BeautifulSoup, Requests
  • Data Processing: Pandas, NumPy for transformation
  • Databases: MySQL, MongoDB for storage
  • Data Lake: Golden dataset repository
  • Job Orchestration: Scheduled jobs for subscriber feeds
  • Monitoring: Logging, error tracking, data quality checks
  • API Layer: RESTful endpoints for subscriber access

Generative AI Image Tools

Fine-tuned SDXL and ControlNet models using LoRA, cinematic prompts, and BLIP2 auto-captioning pipelines with ComfyUI. Also includes FLUX-1-dev image generation with GPT-4 prompt refinement.

Key Features:

  • Fine-tuned SDXL and ControlNet models with LoRA
  • FLUX-1-dev image generation with GPT-4 prompt refinement
  • Cinematic prompt engineering
  • BLIP2 auto-captioning pipelines
  • ComfyUI integration for advanced workflows
  • Multi-platform support (Apple Silicon, CUDA, CPU)
  • GPU pod optimization (RunPod with memory-saving tweaks)
  • Amazon S3 integration for image uploads
  • Batch processing via Google Sheets
  • Docker containerization for easy deployment

Technical Stack:

  • Models: SDXL, ControlNet, FLUX-1-dev via Hugging Face Diffusers
  • Prompt Enhancement: OpenAI GPT-4
  • Infrastructure: RunPod Serverless, Docker
  • Storage: Amazon S3
  • Batch Orchestration: Google Sheets API
  • Workflow: ComfyUI, BLIP2 auto-captioning

Get In Touch

I'm always interested in discussing new opportunities, challenging projects, or innovative ideas.

souvik79@gmail.com
+44 7442 633106
Basingstoke, United Kingdom
LinkedIn Profile
GitHub Profile