Delhi, India — available for work

Madhav Yadav

AI/ML Intern • SDE Intern. I turn models into real products — LLM tooling, retrieval systems, and full-stack platforms, built to be reliable and fast.

View work Get in touchGitHubLinkedIn
Scroll↓

01 / About

I'm an AI/ML and software engineer who likes turning models into things people can actually use — LLM tooling, retrieval systems, and full-stack apps — with an eye for real-time performance, reliability, and clean engineering.

Delhi, India · AI/ML Intern • SDE Intern

02 / Work

Experience

Oct 2025 — Jan 2026

ML & Software Engineer Intern — Agnirath Aerolabs

  • Architected and deployed a real-time Ground Control Station (GCS) for UAV monitoring and autonomous mission management, integrating telemetry pipelines with sub-second latency.
  • Built high-performance FastAPI backend services in Python to ingest, process, and relay drone flight data and command payloads across embedded UAV systems.
  • Developed responsive Next.js operator dashboards visualizing real-time drone telemetry — position, altitude, battery, and mission status — reducing operator response time significantly.
PythonDjangoComputer VisionTransformersNext.jsUAVs
Sept 2023 — Present

B.Tech in Computer Science — AKTU

03 / Projects

Selected work

Featured

Open–Papertrade

An AI-powered paper-trading platform to practice trading strategies with real market data and zero financial risk.

  • Engineered a full-stack paper-trading platform simulating real-world market conditions, letting users test algorithmic strategies risk-free before live deployment.
  • Built a robust REST API backend with Django for order management, portfolio tracking, and real-time market-data ingestion, supporting concurrent user sessions.
  • Designed for extensibility — the architecture supports future integration of ML-based trading-signal models and backtesting engines.
DjangoNext.jsTypeScriptREST APIs
Project 02

Github MCP Server

A lightweight Model Context Protocol (MCP) server built with FastMCP to expose and automate GitHub operations using LLM-compatible tools.

  • Retrieve a GitHub user's bio.
  • Recursively fetch all files and directories from a repo starting at any path.
  • Fetch the content of a file from a repository.
  • Create a new branch and then open a pull request on GitHub.
  • Open an issue on a GitHub repository.
PythonFastMCPGitHub APILLM Tools
Project 03

Steganographic Encoder/Decoder

A Django web app that lets users hide and retrieve secret messages within images — a practical introduction to steganography.

  • Encode hidden messages into images.
  • Decode secret messages from images.
  • Friendly web interface built with Django.
PythonDjangoPillowOpenCV

04 / Skills

Toolkit

Core LLM / Applied AI

Large Language Models (LLMs)Prompt EngineeringContext EngineeringRetrieval-Augmented Generation (RAG)Agentic RAGHybrid SearchRerankingVector Search / Semantic SearchEmbeddingsAI AgentsAgentic WorkflowsTool Use / Function CallingModel Context Protocol (MCP)Structured OutputsMultimodal AIFine-tuningLoRA / QLoRARLHF / DPOModel Distillation

Evaluation & Quality

LLM Evaluation (Evals)LLM-as-JudgeOffline & Online EvaluationRegression Testing for LLMsGuardrailsHallucination MitigationPrompt Injection DefenseResponsible / Safe AIOWASP LLM Top 10

AI Infra / Platform

LLM Inference OptimizationModel ServingvLLMTGISGLangKV CacheContinuous BatchingQuantization (GPTQ/AWQ/GGUF, INT4/INT8/FP8)Speculative DecodingGPU OptimizationVector Databases (pgvector, Qdrant, Weaviate, Milvus)ANN Indexing (HNSW/IVF)

LLMOps / MLOps

LLMOpsMLOpsObservability / Tracing (Langfuse, Phoenix, Braintrust)Prompt VersioningSemantic CachingPrompt CachingModel Routing / Gateway (LiteLLM)Cost & Latency OptimizationModel MonitoringCI/CD for ML

Frameworks & Libraries

LangChainLangGraphLlamaIndexDSPyHugging Face TransformersPyTorchpydanticFastAPIAnthropic / OpenAI SDKssentence-transformers

Languages & Engineering

PythonTypeScript / JavaScriptSQLasync / concurrent programmingREST APIsGitSystem DesignDistributed Systems

Cloud & Deployment

AWS / GCP / AzureDockerKubernetesServerlessGPU ComputePostgresRedisCI/CD

ML Foundations

Machine LearningDeep LearningNeural NetworksTransformersNLPModel Evaluation MetricsRegularization (L1/L2, weight decay)

05 / Contact

Let's build something worth shipping.

Open to roles and collaborations in applied AI and platform engineering. Email is the fastest way to reach me.

Say helloDownload résumé
Emailmymadhavyadav07@gmail.com
GitHub/mymadhavyadav07
LinkedIn/mymadhavyadav07
Portfoliomymadhavyadav07.vercel.app