Master of Science in Computer Science
May 2026University of Arkansas · Fayetteville, ARGPA 3.75 / 4.0
Image ProcessingPrivacy Enhancing TechnologyFull Stack Deep LearningMachine Learning
About
I am a full stack software engineer with more than three years of professional experience. I build web applications with Python, FastAPI, React, Next.js, and PostgreSQL, and I have hands on experience integrating AI features such as retrieval augmented generation, AI agents, and MCP tooling, testing them for reliability, and deploying them on AWS. My two most recent projects, AdvisorDesk and SentinelBrief, are both live, and you can try them from the projects section below.
I hold a Master of Science in Computer Science from the University of Arkansas. During my research years I worked on reinforcement learning for fuzzing and on anomaly detection in cyber-physical water systems, and that work led to two peer reviewed publications. Alongside the research I built the software around the models, including data pipelines, APIs, dashboards, and tooling.
This combination is what I bring to full stack and forward deployed engineering roles. I can take a customer problem, design the system, and ship it in React, Next.js, Node, or Python. When the problem calls for it, I can also build and evaluate the machine learning behind it.
02 peer-reviewed publicationsBest Student Paper nominee · VEHITS 2024MS Computer Science · GPA 3.75/4.0
Skills
Languages
- Python
- TypeScript
- JavaScript
- SQL
- C
- PHP
- HTML/CSS
Backend & APIs
- FastAPI
- Node.js
- Pydantic
- SQLAlchemy
- REST APIs
- OpenAPI
- Server-Sent Events
- OAuth 2.1
Frontend
- Next.js 16 (App Router, RSC)
- React 19
- Redux Toolkit
- Material UI
- Tailwind CSS
AI & ML
- LLM integration
- RAG
- AI agents
- MCP
- PyTorch
- Reinforcement Learning
- Graph Neural Networks
Data & Cloud
- PostgreSQL
- pgvector
- Supabase
- MySQL
- Redis
- ARQ
- AWS (EC2, ECR, S3, SSM, VPC)
- Docker
- Caddy
Testing & DevOps
- pytest
- mypy
- Vitest
- ESLint
- GitHub Actions
- CI/CD
- Git
- Linux
Education
Bachelor of Technology in Information Technology
May 2017College of Engineering Roorkee (COER) · Roorkee, India
Projects
AdvisorDesk: AI-Powered Advisory Content Platform
Next.jsReactTypeScriptFastAPISupabase (Postgres + pgvector)OpenAIMCPOAuth 2.1DockerAWS
A publishing platform for financial advisory content, with an admin CMS, a client portal, and an AI assistant that answers with source citations.
93% mean source support on an 80 question evaluation, 733 ms median time to first token in production
SentinelBrief: AI Security Alert Triage Platform
PythonFastAPIPostgreSQLRedisARQNext.jsTypeScriptOpenAIDockerAWS
Turns live SSH honeypot sessions into prioritized security alerts with severity, reasoning, and recommended actions, backed by a measured evaluation harness.
About $0.0012 per alert and 1.1 s median model latency across 252 evaluation sessions
Reinforcement Learning Guided Fuzzing
PythonReinforcement LearningFuzzingSecurity Tooling
An RL fuzzer that mutates multiple input parameters per round, reaching 70% code coverage and twice the crashes of traditional fuzzing. IEEE CSR 2025.
70% code coverage and twice the crashes of traditional fuzzing, published at IEEE CSR 2025
Anomaly Detection in Cyber-Physical Water Systems
PythonPyTorchGraph Neural NetworksTime-Series MLEnsembles
GNN based anomaly detection for water infrastructure, reaching an ensemble Oracle F1 of 0.854 on SWaT, with an operator dashboard and LLM explanations.
Ensemble Oracle F1 0.854 on SWaT, above the published GDN baseline
RL for Emergency Vehicle Traffic Optimization
PythonPyTorchStable-Baselines3SUMOReinforcement LearningV2V
Reinforcement learning lane selection that cut emergency vehicle traversal time by up to 10.9% in SUMO. Best Student Paper nominee, VEHITS 2024.
Best Student Paper Award nominee, VEHITS 2024
City Complaint Management System
PHPWordPressJavaScriptHTMLCSSMySQL
A complaint management system for Roorkee's municipal body, built in PHP, JavaScript, and MySQL on WordPress, routing about 300 citizen complaints a month.
About 300 citizen complaints a month routed to the responsible departments
Generative Image Synthesis with Diffusion Models
PythonPyTorchDiffusion ModelsU-NetAutoencoders
DDPM texture synthesis on ALOT and latent diffusion models for text-to-image generation, evaluated on CelebA-HQ and LAION with FID and IS.
Deep Learning Image Steganography
PythonPyTorchComputer VisionEncoder-Decoder Networks
A deep learning framework that hides a full image inside another with high imperceptibility and robust recovery of the hidden data.
Real-time Pedestrian Detection and Safety Alert System
PythonYOLOMobileNet SSDTensorRTDeepStreamNVIDIA Jetson
Real-time pedestrian detection on the NVIDIA Jetson Nano with YOLO and MobileNet SSD, optimized with TensorRT, driving audio-visual safety alerts.
Experience
Graduate Research Assistant
Jan 2024 – May 2026Cybersecurity Lab, University of Arkansas · Fayetteville, AR · Advisor: Dr. Qinghua Li
- Designed the core algorithm of a reinforcement learning fuzzer that mutates multiple input parameters per round. It reached 70% code coverage and twice the crashes of traditional fuzzing on a power generation control system. This work was published at IEEE CSR 2025.
- For my master's thesis, I built graph based and ensemble anomaly detection models for the SWaT water treatment testbed (51 sensors across 6 stages, 35 labeled attacks). They reached a mean F1 of 0.886 (SD 0.009) across five seeds, against a published benchmark of 0.81.
- Made model training reproducible with orthogonal embedding initialization, gradient clipping, and cosine annealed learning rate warmup. Failed training runs dropped from 40% to 2%, and the F1 standard deviation fell from 0.066 to 0.020.
Case studies: Reinforcement Learning Guided Fuzzing · Anomaly Detection in Cyber-Physical Water Systems · Generative Image Synthesis with Diffusion Models
Research Assistant, RL for Traffic Optimization
Jun 2022 – Dec 2023IIIT Hyderabad · Hyderabad, India · Advisor: Dr. Praveen Paruchuri
- Trained reinforcement learning lane selection policies in Python and PyTorch on a 2 km four lane network in the SUMO traffic simulator. They cut emergency vehicle traversal time by up to 10.9% against the strongest of three heuristic baselines, evaluated over 100 runs per setting.
- Benchmarked the learned policy against SUMO's default lane changing, a fixed lane policy, and a utility weighted best lane heuristic, using an empty road run as the theoretical lower bound.
- Validated the policy against New York City traffic speed data and wrote the peer review response defending the method. The paper was accepted at VEHITS 2024 and nominated for the Best Student Paper Award.
Case study: RL for Emergency Vehicle Traffic Optimization
Software Engineer
Jun 2021 – Apr 2022Codeventure Tech LLP · Roorkee, India
- Launched a complaint management system for Roorkee's municipal body in PHP, JavaScript, and MySQL on WordPress. It routed about 300 citizen complaints a month to the responsible departments.
- Built the citizen facing reporting interface with client and server side validation, so complaints reached department queues complete and staff no longer had to contact citizens again for missing details.
- Designed the relational schema for complaint records and their handling states, from submission through routing to resolution, and trained administrative staff on the response workflow at handover.
Case study: City Complaint Management System
Software Engineer
Aug 2017 – Mar 2020CETPA Infotech Pvt. Ltd. · Roorkee, India
- Delivered PHP business logic, REST APIs, and form workflows for 4 client web applications, taking each from prototype into production.
- Built more than 15 responsive client websites in HTML, CSS, and JavaScript with custom PHP on WordPress for clients across 3 industries, scoping features directly with each client.
- Extended MySQL schemas and query logic as client requirements changed, and supported clients after launch with fixes and feature updates.
Publications and talks
Uwibambe, M.L., Tyagi, A. and Li, Q. (2025). A Reinforcement Learning Approach to Multi-Parametric Input Mutation for Fuzzing. 2025 IEEE International Conference on Cyber Security and Resilience (CSR), pp. 174–179.
Tyagi, A., Lowalekar, M. and Paruchuri, P. (2024). Improving Lane Level Dynamics for EV Traversal: A Reinforcement Learning Approach. International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS), pp. 134–143.
doi:10.5220/0012637200003702Best Student Paper Award Nominee
Technical talk (Sep 2026). Beyond RAG: Building an Evaluated, Self-Improving Agentic AI System. Charlotte, NC.
Contact
I am open to full stack and forward deployed engineering roles. The fastest way to reach me is by email.