SHREYAS DEB / PORTFOLIO
// HELLO_WORLDI'M
Undergraduate researcher & developer working across AI, systems biology and the low-level guts of computers.
MS DUAL DEGREE · COMPUTATIONAL NATURAL SCIENCES
FOURTH YEAR · IIIT HYDERABAD
Alwaysopentolearningnewstuff.Ifit'scomplex,I'minterested.
An MS Dual Degree student in Computational Natural Sciences at IIIT Hyderabad, where my world revolves around the beautiful chaos of computing. I work as an Undergraduate Researcher under Professor Vinod P K, exploring the intersection of systems biology and deep learning.
A builder at heart. Physics & mathematics are my playground. My interests span the whole computing spectrum — from the low-level logic of kernels & operating systems to the high-dimensional latent spaces of deep learning.
When I'm not debugging a segfault or training a model, you'll find me gaming, watching anime, on the basketball court, or — inevitably — hopping to yet another Linux distro.
Texperia
uExcelerate
MontyCloud
CloudNuro.ai
Center for Computational Natural Sciences and Bioinformatics (CCNSB)
E-Cell IIIT Hyderabad
Open Source Developers Group (OSDG)
IIIT Hyderabad
Atlassian
Selected as a winner among 300+ eligible candidates and received a Pre-Placement Interview (PPI) offer.
IIT Bombay
Selected as Top 25 finalist from 3000+ entries at Action Plan — the flagship social entrepreneurship event of Abhyuday, IIT Bombay's Advitya Social Fest.
HackIIIT
Won the Most Impactful award at HackIIIT 2025 for Saathi, an AI-powered mental wellness platform.
IIIT Hyderabad
Awarded Dean's List III recognition for exceptional academic performance (top 15%).
IIIT Hyderabad
Recognized on the Merit List for outstanding academic achievement (top 20%).
FROM KERNELS TO MODELS — THE WHOLE STACK08 DOMAINS / 61+ TOOLS
KEEP SCROLLING — THE TRACK MOVES SIDEWAYS
Two fully independent, from-scratch decoder-only Transformer language models for Hindi and Nepali, built across three phases: data pipeline + tokenizer, pretraining + evaluation, and reasoning finetuning + attention analysis. No shared data, vocabulary, or weights between the two languages, and no pretrained models, tokenizers, or HuggingFace transformers classes anywhere; positional embeddings, multi-head attention, and causal masking are all implemented from scratch. Includes a custom BPE tokenizer swept across 5 vocab sizes, a KenLM-based perplexity filter for corpus cleaning, and a synthetic reasoning dataset generator for finetuning.
Hybrid BM25 + dense retrieval search engine built from scratch (no IR library dependency), with a LambdaMART learned reranker and LLM query expansion on top. Validated end-to-end on BEIR/SciFact against real relevance judgments — each stage (bm25 < dense < hybrid < rerank) shows a clean, honestly-measured improvement, including an A/B test where query expansion was found to hurt more than it helped.
Event-driven, temporal enterprise knowledge graph with an agentic workspace on top. Watches an issue tracker and wiki via webhooks, extracts entities with an LLM, resolves them against a canonical graph, and writes bi-temporal facts to Neo4j — every edge knows both when it became true and when Apollo learned it. An MCP server enforces an unbypassable permission filter over every graph query.
Modular statistical arbitrage engine for NSE equities. Cointegration-based pair selection (Engle-Granger plus PCA/DBSCAN clustering), a rolling-OLS or Kalman-filter spread model, z-score or XGBoost-gated signals, and a bar-by-bar backtester with transaction costs — exposed via CLI, an interactive TUI, and a set of guided notebooks, backed by 112 tests.
A ground-up reimplementation of Apache Kafka in Go: a distributed pub-sub message broker backed by a Raft-based controller quorum (KRaft-inspired) with automatic leader election, append-only partitioned commit logs, and consumer-group coordination. Ships a full admin CLI for topic management, producing, and consuming — built directly from the original Kafka and Raft papers.
Unix shell implementation in C with integrated Gemini AI capabilities. Features include natural language command translation, command explanation, AI chat, custom readline implementation using termios (no ncurses), 40+ built-in commands, job control, background process management, and tab completion. Built entirely with POSIX APIs and raw OpenSSL sockets for HTTPS without external dependencies.
