Sourab Shaik.

1. Software Engineer

   Dec 2025 – Present

   I build client applications across backend services and frontend interfaces, with responsibility for cloud deployment and model evaluation.

   • Autonomous Meeting Scheduling Agent

     For a Microsoft 365 email scheduling agent, I kept scheduling rules in code and built the conversation workflow with LangGraph. The agent preserves conversation state across restarts and includes an approval step in Teams.

     My work also covered booking intent, invitee selection, and replay handling, plus evaluation gates, deployment infrastructure, and token accounting for an internal pilot.

     Python, LangGraph, Azure OpenAI, Cosmos DB, Service Bus, Microsoft Graph, Bicep

   • Clinical Registry Analytics Platform

     I built tools for exploring a clinical registry and testing hypotheses: a graph explorer with ontology mappings and field explanations, configurable dashboards, and a statistical analysis service.

     I implemented Cox proportional hazards, Kaplan–Meier, and Aalen–Johansen estimators for survival analysis. I also corrected prevalence scaling and comorbidity encoding that had produced misleading results.

     Python, FastAPI, Neo4j, Next.js, TypeScript, Docker

   • Enterprise Document Assistant

     Employees can ask questions about an internal procedure library in the application or in Teams, then inspect the source documents in an embedded PDF viewer. I built the retrieval-augmented generation application behind that workflow.

     I enforced document permissions during retrieval and used SharePoint webhooks to update the search index when documents change. I also added semantic reranking.

     Next.js, TypeScript, Azure AI Search, Vercel AI SDK, Entra ID, Microsoft Graph

   • AI Gateway

     I took a LiteLLM prototype through CI setup, deployment infrastructure, and production cutover. I documented the request pipeline and specified the control plane for the next phase.

     LiteLLM, Railway, GitHub Actions, Blacksmith

   • Distributed Tracing and Document Processing

     I designed a document-processing job lifecycle and versioned Go API, then led tracing across Python, Go, and Next.js so requests could be followed through background messages and workflows.

     I added structured logs and Grafana dashboards as code, with sampling and log-cardinality policies to control telemetry volume.

     Go, Python, TypeScript, GCP, Terraform, OpenTelemetry, Grafana

2. Data Science & AI Engineering Intern

   Jan – May 2025

   I cleaned industrial process data and used principal component analysis (PCA) and clustering to investigate operating conditions associated with higher efficiency.

3. Software Engineering Intern

   Feb – Aug 2024

   I developed Next.js and Flutter applications with REST APIs. For a grant-writing application, I built a LangGraph agent that used client documents to generate grant narratives.

   • MDKCopilot · Race telemetry

     I built software to turn live race telemetry into coaching cues for drivers. During IMSA race weekends, I supported the application trackside and debugged problems between sessions.

     Read BGSU’s feature on the project.

     Python, real-time telemetry

1. Network Assistant · ITS

   Oct 2021 – Dec 2025

   I installed and maintained campus networking equipment and troubleshot connectivity problems.

2. President · ACM Student Chapter

   Aug 2022 – May 2024

   I organized workshops and hackathons for the computer science community, along with social events and industry collaborations.

Completed honors project · 2025

Explainable AI for
liver transplantation

I compared Cox proportional hazards, Random Survival Forest, and DeepHit to test whether molecular HLA mismatch features improve five-year liver transplant outcome prediction. I used matched cross-validation and shared preprocessing, defining all-cause graft failure as the earlier of graft failure or death.

In this cohort under the pooled endpoint, molecular HLA features did not materially improve discrimination over clinical baselines. DeepHit showed a small, variable gain. Clinical variables dominated the explainability summaries, while DR-locus physicochemical metrics showed modest signals.

CPH · Random Survival Forest · DeepHit · Model-agnostic explainability
Advisors: Dr. Robert Green · Dr. Christopher Rump

Read the honors project on ScholarWorks@BGSU ↗

Academic projects

• ASL Recognition System

  I built an ASL recognition tool for learners using TensorFlow Lite and OpenCV.

  Python, TensorFlow Lite, OpenCV

• Honors Project Tracker

  I built a Flutter application for students and faculty to track Honors College project milestones, deadlines, and requirements.

  Flutter, Dart

• Ethical AI Analysis

  I examined ethical questions in AI deployment and proposed guidelines for responsible use.

• Strafe · Keyboard-driven web search for macOS

  I built an open-source macOS app for searching the web and reading results without leaving the keyboard. Press Option + Space from any app, search, move between articles with Vim-style keys, and return to your work with Escape.

  Strafe fetches pages in parallel and turns them into clean reader views with Mozilla Readability. The website includes a 37-second demo and screenshots of the reader, image, and raw-page views.

  Tauri, Rust, TypeScript, Mozilla Readability, Brave Search API

  View Strafe & demo ↗Source on GitHub ↗

• Job Search Tool

  In development.

  I designed a knowledge base for resume and cover-letter generation. It separates profile data from experience records and preserves the source of each accomplishment, with snapshots of the records used for a generated document.

  I built a prototype to evaluate models and defined the generation agents’ input and output contracts. I have not built the production generation backend.

  TypeScript, Python, PostgreSQL, JSONB, OpenRouter

• AI-Assisted Engineering Workflow

  I wrote agent skills for planning and implementation, including worktree coordination and pull request review. I use them while gathering requirements, making architecture decisions, and checking the resulting code.

  Codex, Claude Code, Git worktrees, GitHub CLI, GitHub Actions

Languages

Python, TypeScript, Go, JavaScript, SQL, Cypher, R, C++, Dart

AI systems

LangGraph, LangChain, Azure OpenAI, RAG, Azure AI Search, LiteLLM, OpenRouter, model evaluation

Backend & data

FastAPI, Pydantic, Node.js, PostgreSQL, Neo4j, Cosmos DB

Web & mobile

React, Next.js, Tailwind CSS, NextAuth, MSAL, Flutter

Cloud & delivery

Azure, GCP, Railway, Docker, Bicep, Terraform, GitHub Actions

Observability

OpenTelemetry, W3C trace context, Grafana, structured logging, dashboards as code