SWASTIDIP MAJI
AI/ML Engineer & Full-Stack Developer | Data Science Undergraduate
Swastidip Maji is a proactive and innovative Computer Science and Engineering student specializing in Data Science with a strong focus on AI/ML. With a solid foundation in deep learning, natural language processing, and full-stack development, he has engineered and deployed multiple impactful SaaS platforms and AI applications. He excels in building robust, scalable solutions, demonstrated through projects like an AI-powered portfolio builder, a cloud cluster detection system, and a multi-agent legal assistant. His technical leadership extends to significant open-source contributions and a ...
Tech Stack
Tools and technologies I use to bring ideas to life
Languages
AI/ML Frameworks & Libraries
Programming Languages
AI & LLM Engineering
ML & AI Frameworks
AI/ML & LLM Engineering
Web Technologies & APIs
Deep Learning & Computer Vision
Natural Language Processing & LLMs
Machine Learning & Deep Learning
LLM Engineering
ML & Modeling
APIs & Interfaces
NLP & LLMs
Data Science & Analytics
Development Tools & DevOps
Cloud & DevOps
Data & Analytics
Web Development & APIs
Data Engineering & Analytics
Web Frameworks & APIs
Data Analysis & Tools
Databases
Cloud & MLOps
Featured Projects
A selection of my recent work and open-source contributions
Portfolio AI – AI-Powered Portfolio SaaS Platform
Developed a multi-tenant SaaS platform that transforms resumes into live, cyberpunk-themed portfolio websites leveraging Gemini-powered PDF parsing with hybrid OCR and metadata extraction. Engineered a config-driven architecture with dynamic subdomain routing via Next.js Middleware and Vercel Domains API, enabling instant portfolio deployment. Implemented a full SaaS backend on Supabase with Google/GitHub OAuth, LemonSqueezy payments, custom domain provisioning, and a RAG-powered chat widget ...
ISRO BAH 2025 – Tropical Cloud Cluster Detection
Built a CNN-based Tropical Cloud Cluster detection system capable of processing over 500MB of HDF5 satellite imagery with 99%+ accuracy. Architected an automated pipeline that processes half-hourly INSAT files with real-time monitoring and CSV/JSON exports. Deployed a Next.js dashboard on Vercel for side-by-side visualization, geospatial mapping, and temporal filtering UI.
MedXpress – Healthcare SaaS Platform
Developed an end-to-end healthcare SaaS platform featuring appointment booking, medication ordering, and AI-powered health insights. Architected a REST API handling over 40 endpoints for appointments, medications, lab reports, and AI health recommendations. Implemented an advanced medication catalog search with fuzzy matching, synonym expansion, and caching, achieving sub-100ms response times.
Sentinel AML Copilot – Money-Laundering Anomaly Detection
Built a detection engine combining 7 typology rules with an Isolation Forest anomaly model over 205,419 synthetic retail-banking transactions, producing 839 alerts each carrying machine-readable trigger evidence. Measured accuracy against ground truth (precision 0.43, recall 0.08, F1 0.13) after injecting hard negative cohorts. Engineered a medallion pipeline (Parquet → DuckDB → Neon PostgreSQL) with a 12-check data-quality gate that rejects bad batches, loading 231,519 rows with exact decima...
RoadGuard – Real-Time Road Hazard Detection
Trained a YOLOv11n object detector on a custom 6,439-image road-hazard dataset (5 classes), achieving mAP@0.5 of 50.3% at 22 FPS on a single RTX 3050. Cut duplicate reports of the same physical hazard using DBSCAN clustering over GPS coordinates, surfacing deduplicated detections on a Folium map with SQLite-backed history. Implemented automatic face and license-plate blurring before storage.
More Projects
Legal-Eagle-Ultimate – Multi-Agent AI Legal Assistant
Engineered a multi-agent orchestration system coordinating 5 specialized Gemini-2 agents for comprehensive risk assessment and compliance checks. Implemented an InLegalBERT fine-tuned NER pipeline for Indian statute recognition, with section/act retrieval across 1872+ provisions. Architected a document processing pipeline combining pdfplumber, Google Document AI OCR, and RAG vector embeddings for generating advice.
Portfolio AI – AI-Powered Portfolio SaaS (200+ users)
Built a multi-stage PDF-to-JSON extraction pipeline: a pdf-lib annotation scanner recovering hyperlinks invisible to text extraction, a two-library text fallback chain, and Gemini vision constrained by a 5-entity schema, with vision-to-text failover. Shipped a streaming LLM assistant on a multi-tenant serving layer with edge middleware resolving wildcard subdomains and custom domains via a SECURITY DEFINER Postgres RPC, HMAC-verified webhooks for payment providers, and Vercel Domains API prov...
Work Experience
My professional journey and key contributions
Junior ML Engineer
- ▹Own project delivery end to end - shipped 3 production projects since joining, spanning ML workflows, data pipelines and the backend services around them
- ▹Build ML workflows on self-hosted open-source LLMs rather than paid vendor APIs, covering model selection, prompt design, retrieval and evaluation, keeping inference off a per-token cost curve
- ▹Proposed and implemented the infrastructure cost reduction that lowered the company's monthly run rate, replacing paid inference and managed services with self-hosted open-source alternatives
National Finalist (Top 50 / 8,000+)
- ▹Selected among the top 30 of 8,000+ national entries for building a CNN-based Tropical Cloud Cluster detection system.
- ▹Designed an automated pipeline processing half-hourly INSAT files with real-time monitoring and CSV/JSON exports.
- ▹Deployed a Next.js dashboard on Vercel with side-by-side visualization, geospatial mapping and temporal filtering; presented the platform to ISRO scientists at the national finals.
Open Source AI/ML Contributor & Technical Lead
- ▹Built 5 open-source AI/ML applications including a multi-tenant Portfolio SaaS (portfolioai.in), CNN-based cloud detection (ISRO-BAH-2025), and RAG systems (Legal-Eagle), all deployed with live demos and comprehensive documentation.
- ▹Architected production ML pipelines using TensorFlow, OpenCV, FAISS, and Gemini API integrations, with Next.js/Streamlit frontends and automated processing workflows across Python/TypeScript codebases.
Merit Scholarship Recipient
- ▹Awarded competitive merit-based West Bengal Government scholarship (Top 10%) for 3 consecutive years, recognizing sustained academic excellence and performance.
Get in Touch
Have a project in mind or want to collaborate? I'd love to hear from you!
SWASTIDIP MAJI
Open to opportunities
Prefer email directly? swastidip2004@gmail.com