AI Engineer Resume: Harshita Poojary
Resume:
HARSHITA POOJARY
*********.*******@*****.*** 213-***-**** ï harshitapoojary § HarshitaDPoojary Ð HarshitaP @ Publications EDUCATION
University of Southern California, Los Angeles, CA. Aug 2023 - May 2025 Master of Science - Computer Science GPA: 3.81/4.00 University of Mumbai, Mumbai, India. Jul 2015 - May 2019 Bachelor of Engineering - Computer Engineering GPA: 3.83/4.00 TECHNICAL SKILLS
Programming Languages: Python, C++, JavaScript, TypeScript, Go, Java, Scala, HTML/CSS Frameworks & Libraries: Node.js, LangChain, PyTorch, TensorFlow, Keras, OpenCV, PySpark, AngularJS, Streamlit Databases: MongoDB, Redis, MySQL, Cassandra, Elasticsearch, Milvus, DynamoDB, Hadoop DevOps & MLOps: AWS, Azure, Docker, Kubernetes, Jenkins, GitHub Actions, GitLab CI/CD, gRPC, REST APIs Monitoring & Tools: Grafana, Prometheus, Kibana, OpenTelemetry, Git, Jira, Postman, Swagger, PowerBI, Tableau WORK EXPERIENCE
AI Engineer Jul 2025 – Present
Deep Defense Solutions (Remote) USA
• Built Agentic AI pipeline architecture using Python, LangChain, REST APIs, and Elasticsearch vector stores to identify false incident reports, automate routing, and reduce manual verification load for internal operations teams.
• Built guardrail-aware LLM workflows to classify incident reports by severity, credibility, and escalation risk, adding confidence scoring and fallback paths to prevent low-confidence AI outputs from triggering automated decisions.
• Contributing to the content moderation feature by analyzing requirements, documenting and designing REST APIs, building integration tests, and preparing AWS-based deployment workflows to support reliable report classification, routing, and review.
• Collaborated with Engineering, Operations, and Product partner teams to define requirements, plan solution design, align API designs with UX needs, and support iterative feature delivery. Machine Learning Engineer Dec 2021 – Jul 2023
Reliance Jio Mumbai, India
• Reduced face registration latency by 40% by deploying RetinaFace models with TensorRT FP16 compression using Python and C++ on T4/A100 GPUs, enabling real-time computer vision inference across 2 distributed Kubernetes clusters.
• Increased deployment velocity by 50% by automating CI/CD pipelines and DevOps workflows for Linux-based backend services using Docker, Kubernetes, GitLab, and Azure, embedding health checks, monitoring, and rollback workflows.
• Improved production reliability by 20% by instrumenting logs, metrics, and distributed traces with OpenTelemetry to troubleshoot ML service failures and perform root cause analysis, reducing MTTR across distributed deployments.
• Implemented spoof-detection across 20+ sites at 90% accuracy by curating in-house annotated datasets tailored to deployment scenarios, fine-tuning ResNet-32 classifiers, and integrating FastAPI endpoints with real-time drift detection and profiling.
• Supervised 4 interns in dataset curation, model validation, code reviews and design reviews, ensuring reproducibility, adherence to coding standards, and production readiness.
Software Engineer Jun 2019 – Dec 2021
Reliance Jio Mumbai, India
• Architected Node.js microservices and content-delivery REST APIs for JioNews/JioGate enterprise systems serving 100M+ users on Kubernetes with MongoDB, MySQL, Redis, and Elasticsearch, implementing sharded data flows, RBAC-based secure access control.
• Built large-scale distributed ingestion pipelines for real-time data processing with Kafka, Node.js, Go, SQL, Kubernetes, improving throughput 35% across multi-device platforms with high availability.
• Implemented gRPC-based authorization between content microservices and a centralized UserService, validating user access via OAuth/RBAC and securing inter-service communication.
• Reduced API downtime by 15% by troubleshooting, debugging, and resolving production issues across REST APIs and Kubernetes services, implementing integration/regression tests, CI validation, and observability with Grafana, Prometheus, and Kibana. Software Engineer Jan 2019 – Jun 2019
TaksyKraft (Remote) Hyderabad, India
• Developed real-time engagement prediction APIs in Python and SQL, improving targeting accuracy and increasing client ROI by 10%.
• Shortened reporting cycles by 40% by building ETL pipelines and Tableau dashboards, accelerating enterprise decision-making. PROJECTS
Meeting Copilot (FastAPI, WebSockets, LangGraph, ChromaDB, SQLite, Slack API, Gmail API) ©
• Designed and built a multi-tenant Agentic AI platform using FastAPI, WebSockets, LangGraph, ChromaDB, SQLite, Slack API, and Gmail API to process live meeting transcripts, extract verified action items, and deliver automated summaries across enterprise workflows. Unified Semantic Space for Multimodal Retrieval (PyTorch, Python, LanceDB)
• Designed a multimodal retrieval framework aligning BridgeTower text–image embeddings for video-based Q&A, integrating LanceDB for semantic search with ranking and deploying an interactive Streamlit interface. 3. Comparison of Model Performance with Knowledge Distillation and Quantization (PyTorch, Python) ©
• Fine-tuned LLaMA-3-8B with 4-bit quantization and LoRA-based logit distillation, incorporating Chain-of-Thought (CoT) prompting to improve reasoning consistency and benchmarking performance on GSM8K (8-shot), achieving 76.9% accuracy with low inference cost.
Contact this candidate