About

I build AI systems end to end: agent loops with real tools, retrieval I can measure, models that run on the device, and the product people actually touch.

I’m Sandeep (Sai Sandeep Kumar Vijayarao), an MS CS student at Northeastern University in San Jose, California, graduating in 2027. I like the part of AI engineering where a model has to survive contact with a real product.

At Trominosoft I built the orchestration, retrieval and tool-use layers of a multi-model platform, load-tested to 10K+ requests a day, then cut its latency 30% and its hallucination rate 38%. At Northeastern I TA three graduate courses, Computer Vision, Generative AI and HCI, and build the evaluation harnesses behind them.

In 2026 I won the Apple Swift Challenge with Anu, an iOS agent that runs Gemma 4 entirely on the phone. Building on that stack led me upstream, and that has become its own journey: I now send fixes to ONNX Runtime, LiteRT, MLX, Swift Collections, LiteLLM and NVIDIA’s auto-ontology, reading codebases I don’t own until I can fix them with a failing test first.

Languages

  • python
  • typescript
  • swift
  • kotlin
  • sql
  • c/c++
  • bash

LLMs and agents

  • openai, anthropic, gemini apis
  • langchain
  • google adk
  • mcp
  • react agents
  • multi-agent

Retrieval and evaluation

  • rag
  • chromadb
  • bm25
  • rerankers
  • sentence-transformers
  • ndcg@10
  • bertscore

Training and inference

  • pytorch
  • huggingface
  • lora
  • dpo
  • vllm
  • int4/int8
  • litert
  • core ml
  • mlx

Vision

  • opencv
  • yolo
  • mediapipe
  • openclip
  • 3d reconstruction

Backend and cloud

  • fastapi
  • node.js
  • next.js
  • docker
  • kubernetes
  • gcp
  • ci/cd

Experience

Open-source contributor

Google, Microsoft, Apple, NVIDIA, Hugging Face, ml-explore

2026 - present

Upstream fixes across on-device runtimes, model tooling and LLM infrastructure: int16 comparison kernels in Google’s LiteRT, a Core ML cache leak in ONNX Runtime, Gemma 3 RoPE scaling in ONNX Runtime GenAI, a zero-size-array crash in MLX Swift, a B-tree fix in Apple’s Swift Collections, Anthropic stream error mapping in LiteLLM, PII handling in NVIDIA’s auto-ontology, and SDK fixes in RunAnywhere and Diffusers, several already merged. The journey and live list.

Graduate Teaching Assistant

Northeastern University

Sep 2025 - present

Built the evaluation harnesses and lab infrastructure behind three graduate courses, Computer Vision, Generative AI and HCI, running agentic AI, on-device inference and multimodal workloads for 100+ MS students across two semesters.

AI Developer

Trominosoft

Feb 2025 - Aug 2025

Engineered the orchestration, retrieval and tool-use layers of a multi-model AI platform in Python on Docker and Kubernetes on GCP, load-tested to 10K+ requests a day behind zero-downtime updates. Instrumented its request paths end to end, then cut production latency 30% by profiling model serving, and reduced hallucination rate 38% with DPO and LoRA fine-tuning on models already serving traffic.

Artificial Intelligence Intern

Infosys Springboard

Sep 2024 - Dec 2024

Unified BLIP for vision, GPT-3.5 for language, and Google TTS for speech behind async API bridges in one platform, cutting task completion time 50%. Benchmarked INT4/INT8 quantization and LoRA on 7B models via vLLM for 3.2x inference throughput at under 2% quality loss.

Software Engineering Intern

Global Techno Solutions

Jan 2024 - Apr 2024

Designed and delivered a full-stack production portal end to end in React 18, a Node.js REST API, and MongoDB, improving operational efficiency 30%.

Education

MS, Computer Science

Northeastern University, San Jose, California

Sep 2025 - 2027

Specializing in NLP, multimodal AI, AI agents, large language models, computer vision, and responsible AI. Admitted with the International Impact Award and the Silicon Valley Campus, Academic Excellence, and University Partner scholarships.

BE, Computer Science with Honors in AI

R.M.K. Engineering College, Anna University

Oct 2021 - May 2025

Published real-time pose estimation research at IEEE AISP 2024 while an undergrad (DOI). Elected Head Boy, representing 600+ students.

Community

Core team at GDG Sunnyvale, organizing AI and ML developer events in the Bay Area. Contributor to IEEE Young Professionals programs, and active across the Apple, Google, and NVIDIA developer ecosystems. Member of the PMI San Francisco Bay Area NextGen Leaders cohort.

Certifications

Linux Foundation: Kubernetes and Cloud Native, and DevOps Practices and Tools (2024). Google Project Management. CodePath AI Engineering (2026). AWS Cloud Foundations and Google Cloud Foundations.

Work
Thirteen case studies across agents, retrieval, on-device AI and vision
Contact
Open to AI engineering and software roles