About

I build production GenAI and bioinformatics infrastructure at Sanofi Vaccines Canada, and research how large language models fail under adversarial and clinical perturbation. My work spans the full stack of applied ML — shipping LLM and RAG systems into production, building the cloud infrastructure and CI/CD pipelines they run on, and evaluating model robustness before they reach users.

M.Eng in Electrical & Computer Engineering from the University of Toronto (GPA 3.94/4.00) and B.Tech in Electrical and Electronics Engineering from NIT Tiruchirappalli.

PyTorch LangChain RAG LoRA / PEFT FSDP vLLM AWS Kubernetes Terraform CI/CD

Experience

Nov 2025 – Present

Machine Learning Engineer (R&D)

Sanofi Vaccines Canada · Toronto, Canada

  • Cut model deployment time from 4 weeks to 2 days with CI/CD pipelines shipping 20 bioinformatics models to AWS HealthOmics across DEV/UAT/PROD, now used by 100+ research scientists.
  • Built the compliant multi-account infrastructure underneath (ECS Fargate, IAM/OIDC federation, SQS, S3, Lambda) with least-privilege promotion gates, now the reference architecture across Sanofi's R&D platform.
  • Project lead and technical point of contact, directing 3 engineers; own production reliability (retry/DLQ/status propagation, CloudWatch metrics and alarms) and nanobody workflow orchestration.
  • Building an agentic auto-deploy layer that drives the framework end to end, removing the remaining manual promotion steps (in progress).
AWS HealthOmics ECS Fargate IAM/OIDC SQS CloudWatch CI/CD
Jul 2024 – Oct 2025

Machine Learning Engineer

Appfabs (contracted to LexisNexis) · Toronto, Canada

  • Shipped production Text-to-SQL and RAG agents (Spark, LangChain, Hugging Face) with distributed multi-node inference on KubeRay and Docker, released through Jenkins.
  • Cut end-to-end query latency by 50% by reworking data processing pipelines, schema indexing, and high-throughput GraphQL/REST API gateways.
LangChain RAG Spark KubeRay Docker GraphQL
May 2024 – Jul 2024

Applied Research Intern

Vector Institute · Toronto, Canada

  • Fine-tuned LLaMA-3.1 across multi-GPU clusters with FSDP and served it behind low-latency vLLM endpoints.
  • Designed a diagnostic RAG system whose chain-of-verification step held hallucinated clinical entities under 1%.
LLaMA-3.1 FSDP vLLM RAG
Sep 2022 – May 2024

Machine Learning Engineer (Mitacs Fellow)

Scribble Data · Toronto, Canada (concurrent with MEng)

  • Benchmarked LoRA against INT8/INT4 quantization, cutting inference memory overhead at minimal accuracy loss and making edge deployment viable.
  • Built privacy-aware LLM agents for legal document analysis with domain masking and automated PII redaction.
LoRA/PEFT INT8/INT4 PII Redaction
Jul 2019 – Nov 2021

Backend Developer

G2 · Bengaluru, India

  • Built high-throughput GraphQL APIs and async ETL (Rails, PostgreSQL, Sidekiq); cut batch data-export runtime from 72 hours to under 20.
GraphQL Rails PostgreSQL Sidekiq

Publications

Embeddings to Diagnosis: Latent Fragility under Agentic Perturbations in Clinical LLMs

R. K. Vijayaraj · Workshop on Evaluation of Generative AI, KDD 2025

Selected Projects

Eval Harness (SPAR)

Score-based robustness metrics for computer-use agents under prompt injection.

Multimodal Clinical RAG

GPT-4V + ClinicalBERT retrieval with chain-of-verification grounding.

Diagnostic LLM Audit

Diagnostic LLMs fine-tuned on synthetic notes; perturbation-audited for stability.

Education

2022 – 2023

University of Toronto

M.Eng, Electrical & Computer Engineering · GPA 3.94/4.00

2015 – 2019

NIT Tiruchirappalli

B.Tech, Electrical and Electronics Engineering

Skills

ML & GenAI

PyTorchHugging FaceLangChain RAGLoRA/PEFTINT8/INT4 FSDPvLLM

Evaluation & Reliability

LLM EvalsRobustness Auditing Red-teamingGrounding

Systems & Cloud

AWSDockerKubernetes KubeRayTerraformSpark CI/CD

Languages & Data

PythonSQLC++ RubySnowflakeGraphQL RESTPostgreSQL