Skip to content

AI that earns a place in the workflow.

I take ambiguous problems from first principles to dependable products, owning the AI, the engineering, and the hard edges between the two.

View work
View resume
Anshul Ranjan standing on a street in Interlaken, Switzerland.

A lens on the work

Start with the proof you care about.

Three ways into the same body of work. Pick the question you need answered first.

Taking an uncertain brief all the way to something people can depend on.

In focus
Meta-ROS
What it is
A hardware-agnostic robotics operating system, shipped as a Python package.
Domain
Systems and robotics
See Meta-ROS

Now

Updated Aug 2026

At Morgan Stanley, I take ownership of end-to-end systems that help people make faster, more confident decisions. My work spans product engineering, data foundations, and AI capabilities that need to be reliable under real-world pressure.

Outside work, I keep building at the edge of AI and systems engineering: exploring efficient models, shipping open-source tools, and turning research ideas into products people can use. I am looking for a small, ambitious team where I can take ownership from 0 to 1.

Work that held up in production

AI systems, data platforms, and research with a measurable outcome on the other side.

Morgan Stanley

Software Engineer, ML

Jan 2025 to Present

Bengaluru, India

  • Engineered a production RAG pipeline with hybrid retrieval, cross-encoder reranking, semantic caching, and context compression, improving retrieval accuracy and response grounding for financial AI applications.
  • Built an internal LLM evaluation framework that benchmarks multiple foundation models on financial workflows, enabling rapid experimentation, model selection, and regression testing through automated quality metrics.
  • Architected and own an AI trading intelligence platform used by more than 2,000 front-office traders, cutting portfolio rebalance review time by 80 percent through real-time analytics and workflow automation.
  • Reduced cross-source data anomalies by 40 percent and improved operational efficiency by 25 percent with an AI validation platform for automated anomaly detection and real-time quality monitoring.

BaseRock.ai

ML Researcher

Sep 2024 to Jan 2025

Mountain View, California

  • Scaled automated test generation 14x with a self-correcting LangGraph agent, plus evaluation pipelines that benchmarked test quality, correctness, and coverage across enterprise codebases.
  • Cut QA costs by 80 percent, raised developer productivity by 40 percent, and reached 94 percent test coverage by building an LLM evaluation and benchmarking platform with automated experiments and CI/CD workflows.

MUFG

Software Engineer Intern

Dec 2023 to Feb 2024

Bengaluru, India

  • Worked on internal banking systems as part of a cross-border engineering team, shipping backend services against a large existing Java codebase.

Center for Cloud Computing and Big Data, PES University

ML Research Intern

May 2023 to Aug 2023

Bengaluru, India

  • Built a hybrid HMM and deep learning speech recognition system for resource-constrained robots, designing the evaluation metrics and benchmark datasets, and reaching up to 26.59x faster inference on a Jetson Xavier NX.
  • Published the result at IEEE MoSICom 2024 in Dubai.

Selected work

Research implementations, shipped libraries, and the platforms around them.

All projects

Education

PES University

BTech, Computer Science Engineering, specializing in Machine Intelligence and Data Science.

Dec 2021 to Jun 2025 · Bengaluru, India

CGPA
9.74 / 10
MRD Scholarship, 6 semesters
Awarded for ranking in the top 2 percent of the department in every eligible term.
Teaching Assistant, 2 academic years
Data Analytics and MLOps, with hands-on Azure ML pipeline sessions and course materials.
Best Capstone Project Award
For Meta-ROS, which also became a US patent application.
Show earlier achievements

  • CBSE board results

    12th: 94.4 percent. 10th: 94 percent.

  • JEE Main

    All India Rank 19,695 among more than one million candidates.

  • KCET

    Rank 843 among more than 200,000 candidates.

  • BITSAT

    296 / 390.

  • International Mathematics Olympiad

    SOF International Mathematics Olympiad, All India Rank 48.

  • National talent search

    Qualified NTSE Stage 1.

  • National Children’s Science Congress

    First place at district level and a national-level participant in 2016.

  • Technical competitions

    Top 5 at Kodikon, Top 10 at Hallothon, and first place in the CyberSecurity Treasure Hunt.

Skills and expertise

Capabilities first. The tools are there to make the work concrete, not to fill a logo wall.

LLM systems and RAG

Hybrid retrieval, cross-encoder reranking, semantic caching, and context compression, tuned for grounding rather than for benchmark scores.

  • LangChain
  • LlamaIndex
  • ChromaDB
  • FAISS
  • Redis

AI agents

Self-correcting agent graphs with tool use and structured output, built to recover from their own failures instead of retrying blindly.

  • LangGraph
  • PydanticAI
  • DSPy

Model architecture research

Linear-attention and mixture-of-experts variants, implemented from the paper up with no custom CUDA kernels required.

  • PyTorch
  • JAX
  • TensorFlow
  • RWKV
  • LoRA
  • PEFT

Evaluation and benchmarking

Evaluation harnesses that make model selection a measurement rather than an argument, with regression gates wired into CI.

  • MLflow
  • Weights and Biases
  • TruLens
  • pytest

Inference optimization

Latency and memory work on constrained hardware, from batching and quantization down to embedded targets like the Jetson Xavier NX.

  • vLLM
  • Ray
  • ONNX
  • TensorRT

MLOps and platform

Data and deployment platforms that move reliably from raw input to a monitored service, whether they run in the cloud or on one machine.

  • Kafka
  • Spark
  • Airflow
  • Docker
  • Kubernetes
  • PostgreSQL
  • Redis
  • GCP
Beyond the core sixA broader engineering and research toolkit.

Languages

  • Python
  • Java
  • C/C++
  • SQL
  • TypeScript
  • Go
  • Bash
  • R

Data and infrastructure

  • Spark
  • Hadoop
  • Kafka
  • Airflow
  • Docker
  • Kubernetes
  • PostgreSQL
  • Redis
  • MongoDB
  • Microservices
  • APIs

Application and observability

  • Angular
  • React
  • FastAPI
  • Flask
  • Grafana
  • Prometheus
  • Maven
  • Gradle
  • JUnit

ML and research toolkit

  • TensorFlow
  • PyTorch
  • JAX
  • Keras
  • scikit-learn
  • NLTK
  • LangChain
  • LangGraph
  • DSPy
  • PydanticAI
  • LlamaIndex
  • vLLM

Cloud and delivery

  • AWS
  • GCP
  • SageMaker
  • Vertex AI
  • Git
  • Linux
  • CI/CD

Problem domains

  • LLMs
  • RAG
  • AI agents
  • Evaluation
  • Benchmarking
  • Fine-tuning
  • PEFT
  • Embeddings
  • MLOps
  • Distributed systems
  • ASR
  • MCP

Research and patents

Published work in speech recognition and robotics middleware, plus one patent filing.

Full record

  1. 2026

    Preprint

    Meta-ROS: A Next-Generation Middleware Architecture for Adaptive and Scalable Robotic Systems

    arXiv preprint, 2026

    Presents the Meta-ROS middleware architecture, cross-platform communication layer, and benchmarks against ROS 1 and ROS 2, reporting up to 30 percent higher throughput than ROS 2.

  2. 2025

    Published

    Evaluating Robotic Operating Systems: A Survey

    CDSR 2025

    Surveys the robotics middleware landscape and the compatibility problems that motivated Meta-ROS, comparing transport models, message schemas, and platform support across existing systems.

  3. 2025

    Filed January 2025

    Meta-ROS: An Advanced Robotics Operating System

    US Patent Application

    A specialized robotics operating system for AI and ML applications. Also received the Best Capstone Project Award at PES University.

Looking for the next hard problem.

Open to AI engineering and ML research roles, on a small team where research and production are the same job. Remote, or open to relocation anywhere worldwide.

anshulpranjan@gmail.comView resume