Portfolio / 2026

Debayudh Basu

AI/ML & Full-Stack Engineer. B.Tech CS student at RCCIIT building production RAG systems at Webel and shipping Web3 protocols. I work across the AI infrastructure, MLOps, and on-chain layers — the parts where systems either hold up under real load or they don't.

Scroll / 01

Selected Work / 02

Selected work

Team-Scoped AI Knowledge Assistant with hybrid retrieval & MCP integration.

Devask hybrid Pinecone and BM25 retrieval engine with MCP tool generation architecture diagram

Architected a full-stack AI knowledge base from scratch — Go/Cobra CLI + REST API server and a Python/FastAPI AI microservice, containerised with Docker Compose and an optional Caddy reverse proxy.

Go
Python 3.10
FastAPI
Pinecone
Docker
Cobra
BM25
Cross-Encoder
MCP

View source

Real-time Trading Engine with sub-second Pyth Oracle streaming.

DhanSetu real-time SSE streaming and automated execution engine architecture diagram

Architected a high-concurrency engine supporting 95+ instruments, utilizing SSE-based streaming and Pyth Network oracles for sub-second price updates, solving latency gaps in retail paper-trading.

Next.js 16
MongoDB
Pyth Oracle
Go
SSE
TypeScript

Live demo(Source private)

End-to-end MLOps pipeline for FX market prediction & registry serving.

USD/INR LSTM forecasting and MLflow registry deployment architecture diagram

Addressed model decay in volatile FX markets by building an end-to-end MLOps pipeline featuring LSTM networks and automated experiment tracking via MLflow to maintain predictive accuracy over time.

TensorFlow
MLflow
FastAPI
Docker
LSTM
Python

View source

On-Chain Prediction Market & Dispute Resolution.

Dike Protocol on-chain prediction market and recursive branching tree architecture diagram

Engineered a collateralised prediction system with Recursive Branching Tree logic; achieved 100% Foundry test coverage; implemented event-driven Solidity state updates with a reactive Wagmi UI.

Solidity
Foundry
Next.js
Wagmi
TypeScript

Live demo View source

Experience / 03

Experience

Two internships. Current: production GenAI for government services. Prior: on-chain protocol work.

Worked on: GenAINow

DSML / AI Engineer Intern

Webel West Bengal Electronics Industry Dev. Corp. Ltd. · Remote

Mar 2025 — Present

Production RAG for government services across 5,000+ BSK centres.

Documents: 326+
BSK centres: 5,000+
Bengali LLM: Local
ChromaDB: RAG

01Built
A RAG system over 326+ government service documents with semantic chunking, ChromaDB embeddings, response citation, and AI guardrails.
02Shipped
A Bengali fine-tuning set for an Ollama-hosted local LLM, plus a hybrid citizen recommendation system with a full unit test suite.

Worked on: Web302

Oct 2025 — Jan 2026

MERN & Web3 Intern

Stellar Code · Remote

Migrated 3,584 users from a v1 platform to v2 without breaking existing behaviour, while shipping new features and an ERC-721 referral tree.

Users migrated: 3,584
Zero-break cutover: v1 → v2
ERC tree: 721
Peak throughput: 5/min

Built
A multi-tier ERC-721 referral tree that encodes membership levels, with on-chain state synced to PostgreSQL via Node.js/NestJS.
Shipped
A v1-to-v2 platform migration for 3,584 existing users with backward-compatible changes, new feature work, and a Uniswap v4 hook that levies protocol fees at 5 tx/min peak on testnet.

About / 04

Crafting logic into intelligence.

B.Tech CS student at RCCIIT building production RAG systems at Webel and shipping Web3 protocols. I work across the AI infrastructure, MLOps, and on-chain layers — the parts where systems either hold up under real load or they don't. Computer Science undergraduate at RCCIIT with production RAG work at Webel and Web3 protocols at Stellar Code.

The future of computing isn't just generating content; it's architecting systems that can reason through ambiguity and execute with precision.

Stack01 / 32 · GenAI & LLMs

RAG

Academic foundation

RCC Institute of Information TechnologyB.Tech Computer Science: 2023 — 2027; 7.55 / 10
St. Stephen's SchoolISC Class XII Science: 2021 — 2023; 84.5%
St. Stephen's SchoolICSE Class X: 2010 — 2021; 87.0%

Honors / 05

Honors & accolades

Ledger01 / 05

2025

ETHGlobal Delhi

ETHGlobal

Prize Winner ($1,000)

One of the top 10 finalists of the prestigious Ethglobal Delhi edition hackathon for building a decentralised multiversal prediction market

Appointment / 2025

Web3 and Blockchain Lead

Google Developer Group (GDG), RCCIIT

Spearheading developer workshops, architectural sessions, and hackathon teams on Web3 and GenAI.

Leadership

Contact / 06

Let's build something intelligent.

debayudhbasu@gmail.com

Available for Summer & Fall 2026 engineering opportunities across GenAI, MLOps, and Web3 protocol infrastructure.

GitHub LinkedIn