Yajush Srivastava

Looking for AI Engineer or Backend Engineer roles. CS Undergrad who can ship powerful AI systems at production grade, fast and efficiently, without fail.

New Delhi, Delhi, India

@yajush

GitHub X (Twitter) LinkedIn

0 followers

Looking for jobs Looking for internships Open to relocating

🎓 B.Tech Computer Science and Technology at GGSIPU · Graduating 2028

Follow Message

About

I am Yajush Srivastava, an AI and Backend Engineer focused on building reliable, production-ready systems that combine machine learning, LLMs, backend engineering, and modern software development. I enjoy taking ideas from an early prototype to a complete working product, including system design, model integration, APIs, validation, testing, deployment, and reliability. My core strength is being able to move quickly while still keeping architecture, maintainability, and engineering quality in mind. I primarily work with Python, FastAPI, PyTorch, scikit-learn, XGBoost, Pydantic, SQL databases, Docker, Linux, Git, React, Next.js, and TypeScript. My AI work includes machine learning, deep learning, computer vision, LLM systems, agents, structured outputs, tool calling, RAG, local models, evaluation, and multi-provider integrations. On the backend side, I work with REST APIs, SSE streaming, PostgreSQL, SQLite, Supabase, Django, Flask, CI/CD, and deployment workflows. I am especially interested in the engineering around AI models. For me, building an AI product is not only about getting a model to produce a good output. The surrounding system also needs clear interfaces, validation, safety boundaries, error handling, fallbacks, persistence, testing, and observability. I prefer architectures where probabilistic AI components are combined with deterministic software instead of giving the model unnecessary control over the entire system. My projects reflect this approach across different areas. SAGE is a local-first voice AI assistant that combines LLM planning, typed tools, speech processing, risk-based execution, persistence, and extensive automated testing. RouteMinds is a real-time transit and ML system built around GTFS data, XGBoost-based travel-time prediction, routing logic, backend services, and large-scale data processing. PlayGroundAI explores multi-agent LLM interactions, model orchestration, evaluation, real-time SSE streaming, and full-stack deployment. I also contribute to open-source projects. I have worked on OWASP BLT across Django backend and frontend systems, improving contributor analytics, activity tracking, UI behaviour, and chart performance. I later contributed to Tracer Cloud's OpenSRE, where I worked on LLM-agent reliability, malformed tool-response handling, regression testing, and technical design around configurable OpenAI- and Anthropic-compatible model providers. These contributions helped me become comfortable working inside existing production codebases, understanding unfamiliar architectures, and making changes that fit established systems instead of rebuilding everything from scratch. My current work also includes CodePreFlight, a local-first terminal tool for Git workflows and AI-assisted code review. It is designed as a quality-control layer between code generation and Git history, combining repository context, deterministic checks, AI review, provider flexibility, privacy controls, and local verification of model findings. I am also building Jutsu Battle, an offline computer-vision game where webcam-based hand-seal sequences are recognised using PyTorch, OpenCV, and MediaPipe and then passed into an independent combat engine. These projects let me work on developer tooling, local-first AI, computer vision, real-time inference, and software architecture from very different angles. I care strongly about reliability and testing. I have built projects with large automated test suites, regression coverage, CI/CD pipelines, structured validation, and explicit fallback behaviour. I try to make systems easy to reason about by keeping components separated and giving each part a clear responsibility. I also care about privacy and human control, especially in local AI and agentic systems where models may interact with tools, repositories, or user data. Although AI engineering is my main focus, I do not want to be limited to one category such as LLMs, agents, machine learning, or computer vision. I see myself as an engineer who can work across the full AI product stack. Depending on the problem, that may involve training or integrating a model, building the backend, designing APIs, working with data, creating evaluation pipelines, implementing local inference, building a frontend, or setting up deployment and CI/CD. I like environments where engineers are expected to move fast, learn unfamiliar systems, and own meaningful parts of a product. I am comfortable using modern coding agents and AI-assisted development to increase development speed, but I still treat generated code as engineering work that needs to be understood, reviewed, tested, and maintained. My overall goal is to become an engineer who can consistently take difficult AI ideas and turn them into dependable products. I want to build systems that are technically strong without becoming unnecessarily complex, use AI where it genuinely adds value, and ship software that people can actually rely on.

Experience

Open Source Contribution

Tracer Cloud/OpenSRE

Apr 2026 – Jul 2026

Improved LLM agent reliability by handling malformed tool-response payloads and adding regression tests, delivered through merged PR #1140; contributed 10 merged pull requests overall. Contributed to architecture and technical design for custom OpenAI- and Anthropic-compatible LLM providers, covering configurable API endpoints, authentication, model overrides, validation, and backward compatibility, including contributions through technical discussions.
Open Source Contribution

OWASP BLT

Oct 2025 – Jan 2026

Enhanced GitHub contributor analytics and activity tracking across Django backend and frontend systems, improving dashboard functionality and usability. Optimized frontend rendering and chart visualization performance; resolved UI overlap and activity-state issues through 5 reviewed and merged pull requests.

Education

B.Tech

Maharaja Agrasen Institute of Technology · GGSIPU

Computer Science and Technology · Aug 2024 – Aug 2028
12th / Higher Secondary (Class 12)

Seth Anandram Jaipuria School · Central Board of Secondary Education (CBSE)

Physics, Chemistry and Mathematics · Apr 2023 – May 2024
10th / Secondary (Class 10)

Seth Anandram Jaipuria School · Central Board of Secondary Education (CBSE)

Apr 2021 – May 2022

Skills

Tools / apps / platforms

Canva 1 to 2 years
Claude 1 to 2 years
Cursor 1 to 2 years
Docker <3 months
Git 1 to 2 years
Jupyter Notebook 1 to 2 years
Linux 1 to 2 years
MySQL <3 months
OpenAI 6 to 12 months
PostgreSQL <3 months
SQLite <3 months
Supabase <3 months

Languages

English Native / Bilingual Speak · Read · Write
Hindi Native / Bilingual Speak · Read · Write

Projects

PlayGroundAI ↗

Next.js, TypeScript, FastAPI, Python, Groq, Gemini, SSE, Supabase

Built a multi-agent GenAI platform enabling real-time interactions between 2+ AI agents. Integrated Groq and Gemini for orchestration and automated evaluation; engineered REST APIs and SSE streaming for low-latency LLM responses. Deployed full-stack services on Vercel and Render with CI/CD.
RouteMinds ↗

Python, FastAPI, XGBoost, Pandas, GTFS/GTFS-RT, React, TypeScript, MapLibre, Supabase

Built an end-to-end transit routing platform using Python, FastAPI, XGBoost, React, and GTFS/GTFS-RT data. Processed 30.8M real-time transit records across 2,167 routes and 5,646 vehicles through validation and feature-engineering pipelines. Improved travel-time prediction MAE by 5.5% over static schedules using guarded XGBoost inference and fallback logic. Developed reliability-aware graph routing with ETA uncertainty, wait-time, congestion, and transfer-risk scoring; validated with 138 passing backend tests.
SAGE: System Assistant for General Executions ↗

Python, FastAPI, Pydantic, SQLite, Ollama/Gemma, Whisper.cpp, Piper TTS, Electron, React, TypeScript, Vite

Built a local-first voice AI agent with 9 typed tools for developer workflow automation using structured LLM outputs and risk-based execution. Engineered an end-to-end pipeline for speech-to-text, LLM intent planning, Pydantic validation, safety enforcement, tool execution, text-to-speech, and SQLite persistence. Implemented 109 automated tests covering agent planning, tool calling, safety policies, API/CLI flows, STT/TTS providers, diagnostics, persistence, and process supervision.

Courses & certifications

Improving Deep Neural Networks: Hyperparameter Tuning ↗ · Stanford (DeepLearning.ai)
Neural Networks and Deep Learning ↗ · Stanford (DeepLearning.ai)
Advanced Learning Algorithms ↗ · Stanford (DeepLearning.ai)
Supervised Machine Learning ↗ · Stanford (DeepLearning.ai)

🎯 Hobbies & interests

Gaming
Music
Dance
Reading cool tech stuffs which blows my mind

Explore the Broxer community

💬 Q&A Ask about roles, salaries and interviews — and help others. 👥 Circles Join discussions on skills, companies and industries. 📰 Feed Follow updates and insights from the community.