Kapish Yadav
ResumeContacts

Kapish Yadav

Kapish Yadav

Final-year DTU student building full-stack apps and AI/ML systems, with a growing focus on agentic workflows.

LinkedInGitHubX (Twitter)

About

I'm a final-year engineering undergraduate at Delhi Technological University (DTU, Class of 2027), with a strong computer science foundation and hands-on skill across C++17, Python, and TypeScript. I work at the intersection of systems programming, full-stack development, and applied machine learning, building software that holds up under real-world load and testing.

Day to day, I build LLM-integrated products spanning RAG pipelines, autonomous agents, and agentic coding workflows, alongside full-stack applications and backend infrastructure on AWS and GCP. My work ranges from low-level systems in C++ to production-facing AI applications, with a growing focus on agentic AI as my primary field.

Skills

Languages

  • Python
  • TypeScript
  • JavaScript
  • C++ (C++17)
  • React

Systems & Backend

  • Distributed Systems Design
  • REST API Design
  • FastAPI
  • Node.js

Machine Learning & AI

  • LLM API Integration
  • Agent Architecture
  • Prompt Engineering
  • RAG
  • LangChain/LangGraph
  • Vector Databases
  • PyTorch
  • scikit-learn
  • XGBoost

Databases & Messaging

  • PostgreSQL
  • MongoDB
  • Redis

Cloud, DevOps & Testing

  • AWS
  • GCP
  • Docker
  • CI/CD (GitHub Actions)
  • CTest
  • Git

Projects

MiniCache — Distributed In-Memory Cache with Capacity-Aware Sharding

GitHub ↗

C++17 · RESP2 Protocol · LRU/LFU Eviction · Consistent Hashing

Architected a distributed, Redis-like in-memory cache from first principles in C++17 implementing the RESP2 wire protocol (compatible with redis-cli), sharded concurrent storage, and consistent hashing (FNV-1a with 100 virtual nodes/shard) to minimize key relocation. Built a 7-suite CTest automation framework, pluggable LRU/LFU eviction, crash-recoverable persistence (AOF, snapshotting), and leader-follower replication. Benchmarked throughput via redis-benchmark across 4 shards at 11,517 GET ops/sec (p99: 7ms) and 9,711 SET ops/sec.

TabNet Studio — PyTorch Research Workbench & Implementation

GitHub ↗

Python · PyTorch · FastAPI · React / TypeScript · Docker · XGBoost

Implemented 'TabNet: Attentive Interpretable Tabular Learning' (AAAI 2021) from scratch directly from paper equations in PyTorch—including Sparsemax attention, Ghost Batch Normalization, and GLU feature transformers rather than wrapping a library. Built an interactive research workbench with FastAPI backend and React/TypeScript frontend for hyperparameter tuning, live training curves, and Sparsemax attention heatmap visualization across forward passes.

Real-Time Physics Engine

GitHub ↗

C++17 · WebAssembly · SAH-Guided BVH · OpenGL 3.3

Built a rigid body physics engine from scratch in C++17 with zero external physics or math libraries—custom Vec3/Mat3/Mat4/Quaternion math, semi-implicit Euler integration, and analytically correct inertia tensors for spheres and boxes. Implemented 15-axis SAT collision detection across five shape pairs, a sequential impulse solver with Coulomb friction and Baumgarte position correction, and a dynamic AABB tree with SAH branch-and-bound insertion for broadphase—achieving a 5.4x speedup over brute force at 1,000 bodies. Compiled to WebAssembly via Emscripten for a live in-browser WebGL 2 demo, backed by 125 passing GoogleTest cases across 5 suites and CI/CD that auto-deploys the WASM build to GitHub Pages.

Financial Time-Series Ensembles — SPY Direction Forecasting

GitHub ↗

Python · scikit-learn · XGBoost · yfinance · TimeSeriesSplit

Built an end-to-end pipeline to forecast SPY next-day price direction from ~4,000 days of OHLCV data, engineering technical features (lagged returns, rolling volatility, Bollinger Bands, RSI) and training Logistic Regression, Random Forest, and XGBoost models with strictly chronological train/validation/test splits and TimeSeriesSplit cross-validation to prevent look-ahead bias. Evaluated both ML metrics (ROC-AUC, precision/recall) and realistic trading performance (Sharpe ratio, max drawdown) via a simulated long-only strategy. Found that buy-and-hold outperformed every model on a risk-adjusted basis—a result that surfaces genuine insight into market efficiency and the overfitting risk of complex ensembles on noisy financial data, rather than overstating a weak signal.

Work history

Globe Capacitors Pvt. Ltd.

Jun 2025 — Jul 2025

Software Engineer Intern

Engineered backend automation and cloud integrations connecting the company's enterprise ERP platform, relational databases, and Slack for automated reporting and alerts on AWS.

  • •Cut manual administrative effort by 40% across 3 operations teams via automated tracking and notifications
  • •Diagnosed and resolved 30+ production defects on GCP infrastructure, reducing average load times by 25% for 100+ internal users
  • •Automated manual approval routing and governance workflows by designing and deploying Python backend APIs, boosting approval completion rates by 15%

Education

Delhi Technological University (DTU)

2023 — 2027

B.Tech in Electronics & Communication Engineering

Class of 2027. Strong computer science fundamentals in data structures, algorithms, systems programming, and operating systems.

Contacts

Location

Delhi, India

I'm currently open to full-time engineering roles, particularly in backend and distributed systems, and in AI-native software development. If you'd like to collaborate or discuss an opportunity, feel free to reach out.