I'm a Master's student in Computer Science (AI track) at the University of Southern California, focused on building scalable, production-grade AI systems across language, reasoning, and multi-agent environments. I work end-to-end from designing neurosymbolic architectures and training small language models to deploying agentic AI workflows and backend services on the cloud.

I'm especially interested in NeuroSymbolic AI, Reinforcement Learning, Knowledge Graphs, and Agentic AI systems architectures that combine the flexibility of neural models with the guarantees of symbolic reasoning. I care about clean system design, explainability, and translating research ideas into practical, high-impact solutions.

Before USC, I spent 3+ years at Shell as a Software Engineer building enterprise API integration solutions that shipped to production and delivered measurable business value.

🎯 Actively seeking full-time job opportunities for 2027 in ML/AI Engineering and ML/AI Research

Backend & Cloud Engineer Intern @ USC Grand Challenges Scholars Program • Los Angeles, CA • Jul 2026 – Present

• Designing and Developing backend engineering solution for USC GCSP Network Portal for 100+ institutions.

ML Research Engineer Intern @ IRT Lab, University of South Carolina • May 2026 – Present

• Developing a GNN-based knowledge graph encoder to replace an ID-pooling baseline in a Neurosymbolic RL system, targeting improved typed-relation reasoning across constraints, preferences, and entities in editable specifications.
• Contributing to the PALMS (Personal Adaptive Learning Management System) project. Designed the Course Knowledge Graph schema and deployed it on Neo4J and built the OCR + transcript ingestion pipeline that extracts video frames, runs Tesseract OCR on slide content, aligns them with timestamped transcripts, and integrates the output with the User KG to drive personalized adaptive learning.
• Contributing to the development of a Small Language Model for food & nutrition

AI Research Engineer Intern @ USC Marshall School of Business • Los Angeles, CA • Feb 2026 – Present

• Built "Mulholland", a Neurosymbolic multi-agent system simulating high-stakes governance negotiations across 13 autonomous agents that negotiate within deterministic legal constraints. Used a self-hosted Qwen 2.5-7B-Instruct model served via vLLM inside SLURM-scoped jobs on USC CARC HPC and scaled to 100 parallel simulations via SLURM job arrays with per-job temporary lifecycle for reproducibility and cost control.
• Engineered end-to-end Python ETL pipeline analysing 30M+ job postings across 10 organizational culture dimensions; processed 50+ Parquet files with schema normalization and cross-file deduplication.
• Executed large-scale web scraping of 6,200+ European data center facilities; built structured datasets with capacity, operational status, location, and infrastructure details.

AI Engineer Volunteer @ USC GRIDS (Graduates Rising in Data Science) • Feb 2026 – Apr 2026

• Engineered a responsible-AI auditing framework enforcing LLM prompt policy compliance at scale: embedding-based policy violation detection, token-budget guardrails, and RAG context relevance scoring with 100% detection accuracy on Natural Questions and LongBench benchmarks.
• Deployed a FastAPI service for context-engineering safety validation and conducted adversarial red-teaming against policy guardrails.

AI Engineer Intern @ Varlyq Technologies • May 2025 – Jul 2025

• Architected an end-to-end AI-powered interview simulator (Python, GPT-4) with audio recording, speech-to-text, TTS generation, and automated answer scoring; hardened reliability via edge-case handling for exit recognition and role reversal.

Associate Software Engineer @ Shell India Markets Private Limited • Bengaluru • Aug 2022 – Jul 2025

• Delivered 7 API Integration Solutions for Shell Market Hub, generating $130K in operational savings and $195K in annual savings through ≥50% reduction in airport processing time.
• Built AIS-based Salesforce MDM integration ($80K/year savings); implemented Salesforce–Shell Recharge via Azure; migrated MuleSoft to AIS; built Pub/Sub + gRPC Platform Events listener eliminating ~2 months of manual migration effort.
• Automated GitHub attestation across 22,000+ enterprise repositories using Node.js; led an Azure OpenAI & Cognitive Search Virtual Assistant POC for Software Engineering workflows.
• Winner — Shell Business Platform Functional Excellence Award (CoE Ideathon 2024); 3× Star of the Month recognitions; Project Process Hackathon (2025)

Agentic AI pipeline that turns a business website URL into a cinematic video ad — fully automated

• Tech: Python, LangGraph, Ollama (Qwen 2.5), Google ADK, Gemini, Nano Banana Pro, Lyria 3, Veo
• Architecture: Strategist ↔ Critic negotiation loop (actor–critic pattern) producing a validated BrandBrief; downstream tool-using agents handle image generation, ranking, music, TTS, and video composition
• Highlights: Structured-output enforcement via Pydantic schemas, hybrid local/cloud stack (open-source reasoning + managed generation), Supervisor agent for feedback routing

Explainable, hallucination-resistant medical coding + Fraud, Waste & Abuse (FWA) detection

• Tech: Python, GPT-4o, NetworkX, ICD-10 / HCC / CMS rulebooks, FastAPI
• Architecture: Neural extraction (LLM pulls clinical entities → ICD-10 / RxNorm / CPT codes) + symbolic reasoning engine that validates every extraction against a healthcare knowledge graph encoding real medical relationships and CMS billing rules
• Highlights: Produces a CMS-defensible audit trail rather than an opaque confidence score; catches upcoding, unsupported billing complexity, and clinical contradictions; designed for payer workflows where explainability is a compliance requirement

• Comparative Performance of Deep Learning Architectures in Lower Grade Glioma Segmentation — Indian Patent and Design Journal
• Breast Cancer Detection: Comparative Analysis of Machine Learning Classification Techniques — IEEE
• Network Security in Software Defined Networks (SDN) — IEEE

Speaker / Invitee — Applied Machine Learning Conference 2026 (USA), API Days NYC 2024, SciML Workshop @ ICERM Lab, API World 2024, AIDevcon 2025 Bangalore.

Languages:
  Primary:   [Python, Core Java]
  Secondary: [C++, C#, JavaScript, Go, SQL]

ML / AI:
  Deep Learning:     [PyTorch, TensorFlow]
  LLM Orchestration: [LangChain, LangGraph, MCP, Google ADK]
  Serving:           [Ollama, vLLM, FastAPI]
  Data:              [NumPy, Pandas]

Databases & Vector Stores:
  Relational:  [PostgreSQL, MongoDB]
  Vector:      [Qdrant, ChromaDB]
  Messaging:   [RabbitMQ, Azure Service Bus]

Cloud & Infrastructure:
  Cloud:       [Microsoft Azure, AWS]
  Containers:  [Docker]
  HPC:         [SLURM, USC CARC]
  Version:     [Git, GitHub Actions]

Backend & Integration:
  Frameworks:  [FastAPI, .NET Core, Node.js, Express.js]
  Azure AIS:   [Logic Apps, Function Apps, API Management, Key Vault]

• Microsoft Certified: Azure Fundamentals (AZ-900)
• Microsoft Certified: Azure AI Fundamentals (AI-900)
• ISC2 Certified in Cybersecurity (CC)
• Cisco CCNA Routing & Switching — Introduction to Networks
• Green Software Practitioner (Linux Foundation)

• NeuroSymbolic AI — combining neural models with symbolic reasoning grounded in Knowledge Graph ontologies for explainable, hallucination-resistant systems to make systems Trustworthy.
• Reinforcement Learning — classical RL, constraint-grounded RL, typed-relation reasoning, and knowledge-graph-conditioned policies.
• ML / AI Infrastructure — self-hosted model serving (vLLM), SLURM-based HPC workflows, distributed inference, and cost-efficient production ML systems.
• Small Language Models — domain-specific pre-training, custom tokenization, and neurosymbolic grounding for specialized vocabularies.

• Research — NeuroSymbolic AI, Reinforcement Learning, Knowledge Graphs, Multi-Agent Systems, Small Language Models, ML Model Optimization and Infrastructure
• Open Source — Agentic AI frameworks, LLM evaluation & monitoring pipelines, knowledge-graph tooling
• Industry Projects — Production ML systems, LLM-based backend services, cloud-native AI applications
• Mentorship — Career guidance for CS students transitioning into ML/AI

"Neural models generalize, symbols reason, and infrastructure decides whether either ever reaches the real world. I build where all three meet grounded, efficient, and small enough to actually run."

Last updated: July 2026