SATADEEP
DASGUPTA

Software engineer specializing in execution runtimes, workflow orchestration systems, and developer infrastructure. Currently building distributed systems and AI-powered platforms as Solution Architect at Brainium Information Technologies. Experienced in systems programming with Rust, C++, Python, and TypeScript.

$satadeep@dev:~

> SKILLS.json

Languages

Rust
C++
Python
TypeScript
JavaScript
Java
Node.js
React

Technologies

Workflow Orchestration
Distributed Systems
Systems Programming
Backend Engineering
Event Sourcing
REST APIs & GraphQL
Docker & CI/CD
Database Systems
Cloud Architecture
Developer Tooling
Infrastructure Monitoring
Model Distillation
Edge AI & Offline Inference
KV Cache Optimization

Spoken Languages

English
Bengali
Hindi

> EXPERIENCE.log

Solution Architect

Brainium Information Technologies Pvt Ltd

Jun 2024 - Present
- **Solution Architect** (January 2025 - Present)
- Lead architecture design for enterprise software systems and cloud-native applications
- Design scalable backend architectures and distributed workflows for mission-critical systems
- Develop AI-assisted systems for automation and intelligent decision support
- Guide system design decisions including service decomposition, data architecture, and observability strategies
- **Associate Software Engineer** (June 2024 - January 2025)
- Implemented core system components for enterprise applications with focus on reliability and performance
- Solved complex backend engineering problems involving integrations and data pipelines
- Worked on distributed application components and performance-sensitive backend services

Full Stack Developer

IGLOBAL IMPACTS ITES PVT LTD

Jun 2022 - Jun 2024
- Designed and maintained scalable web applications and backend services supporting thousands of concurrent users
- Built and maintained CI/CD pipelines and deployment workflows, reducing deployment time by 50%
- Participated in technical hiring and engineering evaluation processes, contributing to team growth
- Implemented production-grade APIs and backend services with strong performance and reliability requirements
- Collaborated with cross-functional teams to deliver complex features on aggressive timelines

Freelancing Developer

FIVERR.COM

Aug 2019 - Jun 2022
- Delivered end-to-end web application projects for international client base across diverse industries
- Designed full-stack systems from requirement gathering through deployment and maintenance
- Built scalable backend APIs and optimized frontend performance for responsive user experiences
- Maintained production deployments and resolved complex technical issues under tight deadlines
- Implemented modern web technologies including React, Node.js, and cloud infrastructure for client solutions

> PROJECTS/

@aime.ai/renderer-react — Lesson Renderer Component Library

Composable React component library for rendering AIME lesson packs with slide presentation, BlockSuite whiteboard, inline editing, and full i18n/RTL support

ReactTypeScriptBlockSuiteYjsComponent Librarynpm Package
Published composable React component library for rendering structured educational lesson packs. Ships as `@aime.ai/renderer-react` on npm with 569+ weekly downloads and 13 versions. Later adopted as ...
Demo

AIME Lesson Studio — Cross-Platform Desktop App

Tauri + Rust desktop application for teachers to view, edit, and print lessons and assessments — ships on Linux, ChromeOS, Windows, and macOS

TauriRustReactTypeScriptDesktop AppCross-Platform
Cross-platform desktop application built on Tauri + Rust for teachers to view, edit, and print AIME lesson packs. Ships as native installers across Linux (.deb, .rpm, AppImage), ChromeOS/ARM64, Windo...
Code

AIME — Adaptive Instruction & Misconception Engine

AI-powered learning platform generating personalized lessons through interactive workflows with real-time misconception detection

AI SystemsWorkflow OrchestrationInteractive LearningPython
AI-powered learning platform designed to generate and deliver personalized lessons through interactive learning workflows. AIME transforms education by delivering learning through interactive session...

AIME-Reasoner-2B — Offline Reasoning Model

Distilled 2B-parameter reasoning model with Think Cache and Think Book architecture for fully offline AI tutoring on CPU-only devices

Model DistillationKV Cache OptimizationEdge AIOffline InferencePythonLLM Engineering
Compact 2B-parameter reasoning model shaped through multi-teacher distillation from advanced reasoning systems (Llama 4 family, Opus-class reasoning chains). Built for offline-first education — runs ...

> STORIES/

Deep dives into the engineering behind the work

AIME-Reasoner-2B: Intelligence That Works Through Problems

What happens when you stop waiting for connectivity and start thinking locally

The Problem Nobody Talks About

Most AI progress assumes one thing silently: that there is a server somewhere, waiting to respond. For schools in rural India, for field researchers in low-connectivity zones, for students who study after the router goes off — that assumption breaks everything.

AIME-Reasoner-2B was built for the moments when the cloud is not an option.

What It Is

A compact language model shaped through distillation from multiple advanced reasoning systems — including models in the Llama 4 family and Opus-class reasoning chains. It was not trained only on answers. It was trained on *how answers are formed*.

Different reasoning styles were brought together, aligned, and compressed into a single model small enough to run on a laptop CPU, an old Android tablet, or a Raspberry Pi tucked into a classroom shelf.

> RESEARCH & PUBLICATIONS.log

Bridging Analytics and Semantics: A Hybrid Database Approach to Retrieval-Augmented Generation

Zenodo RepositoryAuthors: Debashis Saha, Satadeep Dasgupta

**Abstract**
Recent advances in Large Language Models (LLMs) have highlighted the importance of Retrieval-Augmented Generation (RAG) in improving factual accuracy, context relevance, and reasoning capabilities. However, most RAG pipelines treat data retrieval and semantic reasoning as disjoint processes, leading to inefficiencies in query execution and knowledge alignment.
In this work, we propose a hybrid database approach that bridges analytics and semantics by combining the structured querying power of SurrealDB with the dynamic reasoning capabilities of LLMs through LangChain and tool execution. Our framework enables fine-grained data access, semantic enrichment, and hybrid retrieval strategies that balance symbolic query execution with contextual generation.
We demonstrate how this integration improves interpretability, reduces hallucination, and enhances query efficiency in knowledge-intensive tasks. This work provides a foundation for building domain-adaptive RAG systems that are both scalable and semantically aware, opening pathways for applied AI in research, enterprise knowledge management, and intelligent assistants.
**Keywords:** Retrieval-Augmented Generation, Large Language Models, Hybrid Databases, SurrealDB, LangChain, Tool Execution, Semantic Retrieval, Knowledge Management, Hallucination Reduction, Query Efficiency
**Key Contributions:**
- Novel hybrid approach combining vector search and SQL analytics in RAG pipelines
- Integration of SurrealDB for unified semantic and structured data retrieval
- LangChain and LangGraph orchestration for intelligent tool selection
- Demonstrated 18% improvement in relevance accuracy over vector-only baselines
- Open-source implementation with modular, extensible architecture
Read PaperDOI LinkSource Code

> CONTACT.sh

Let's build distributed systems and solve complex engineering challenges together

Get In Touch

satadeep3927@gmail.com
📱+91 6289877656
📍Kolkata, West Bengal, India

Social Links

GitHub ProfileLinkedIn Profile