AI / ML engineer and researcher. MS Computer Science at George Washington University on the Machine Intelligence and Cognition track. I build multi agent systems, ship production backends, and publish research.
About
01 / aboutI work where research meets shipping. I like problems where you have to invent the approach and then make it survive real users and real bills.
Right now I am a Software Engineering Intern at BotStacks, working across a FastAPI and GraphQL backend, a React and TypeScript frontend, and a shared core library. I have shipped a full support ticketing system, wired observability across three production services, and fixed a live billing bug that was blocking customers.
In parallel I run my own machine learning research, hold a UK design patent, and build multi agent systems for fun. My aim is model level research at a frontier lab.
Selected work
02 / projectsWanderIQ
End to end travel planner orchestrating 22 specialized LangGraph agents through a shared state graph, from intent and budgeting to routing and itinerary synthesis.
CodeGuardian
An 8 agent code review system with full Langfuse observability. Specialized reviewers cover correctness, security, and style while a human stays in the loop.
TurboForge
A wind farm digital twin using Temporal GANs and Transformers to model turbine behavior. Version 4 reproduced real turbine dynamics with high fidelity.
ICM on Llama 3.1 405B
Implemented the Internal Coherence Maximization algorithm on a 405B parameter model, using unsupervised elicitation of coherent beliefs.
swastik-brain
A personal AI second brain that ingests notes, projects, and history into a queryable knowledge system so context follows me everywhere.
GW Job Monitor
An always on GitHub Actions monitor that detects new on campus job postings within seconds and tailors a cover letter per posting with an LLM.
Experience
03 / experienceRemote
Software Engineering Intern
- Built and merged a full customer support ticketing system end to end across API, core, and frontend.
- Wired Sentry error monitoring and PaperTrail log shipping across all three production services.
- Shipped a support widget and added rate limiting to the public LLM chat endpoint.
- Fixed a production Stripe billing bug where webhook failures blocked customer plan upgrades.
- Resolved N+1 query hotspots with GraphQL batching and added coverage to an untested service.
Data Analyst Intern
- Built ETL pipelines and optimized PostgreSQL queries for analytical workloads.
- Ran cohort analyses that turned raw event data into decisions.
Undergraduate Researcher
- Predicted oil and gas subsurface properties using wavelet transforms and machine learning.
- Work resulted in a co authored IEEE publication.
Research & patents
04 / researchAI Based Computer Device to Detect Water Toxins
A registered design for an AI powered instrument that detects toxins in water, granted by the UK Intellectual Property Office with Swastik Bhat as a named proprietor.
- authority
- UK Intellectual Property Office
- design no.
- 6535622
- granted
- 29 July 2026
- class
- Checking and testing instruments
Confound Control in Subject-Independent EEG Based Alzheimer's Detection
A two cohort study (ds004504 and P-ADIC) benchmarking six model families with a full MNE-Python pipeline, an age only baseline, and cross-cohort transfer. Co authored with Nayana B. R.
Subsurface Property Prediction with Wavelet Transforms and ML
Undergraduate research applying wavelet transforms and machine learning to oil and gas subsurface property estimation, resulting in a co authored IEEE publication.
Toolkit
05 / skillsAI & Machine Learning
Backend & Data
Frontend & Product
Infra & Observability
Research & Analysis
Coursework
Education
06 / educationOpen to AI, machine learning, and software roles, research collaborations, and hard problems.