Maanaav Motiramani

University of New South Wales

• Weighted Average Mark (WAM): 78.9 | Specialization: Computational
• Recipient of UNSW International Scientia Scholarship (A$40K), awarded for outstanding academic merit & leadership.

Relevant Coursework: Machine Learning & Data Mining, Database Systems, Big Data Management, Data Visualisation, Empirical Methods for Causal Inference, Statistical Inference, Foundations of Computer Science, Data Structures & Algorithms, Multivariate Analysis, Business Economics, Economics of Strategy, Information Systems Consulting

SVKM's NMIMS Mukesh Patel School of Technology Management & Engineering (MPSTME)

• CGPA: 3.91/4 | Honours in Data Science/Analytics
• Awarded Technical Achievement Award (1 of 3 students for research & innovation), Meritorious Student Award (Dean's List) (Ranked Top 12 in graduating batch) and selected as Student Ambassador (1 of 8 students) for contribution to institute representation.

• Built an agentic memory pipeline (A-MEM) to classify, save, reinforce, or skip memories in real time, with guardrails for importance thresholds, save rate limiting, and low-signal turn detection.
• Designed a 3-embedding retrieval architecture trait, video and chat context stored in MongoDB Atlas Vector Search, with parallel search legs and weighted score merging, extended by a multi-hop BFS traversal across a bidirectional memory graph.
• Built a study signal pipeline that logs behavioural signals per video progress bucket with age decay and usage weighting, triggering a background pattern detection job that consolidates recurring habits into long-term memories - serving personalised tutoring to 2,000+ users with memory context injected into every response.

• Predicting Mars dust storm intensity from pre-storm atmospheric conditions by merging storm catalog (MDAD) with dust opacity climatology (Montabone CDOD) across 5 Mars years.

• Built an AI question-generation platform for a Monash University course ahead of a pilot with ~350 students: hybrid RAG (embeddings, full-text & RRF on Supabase, OpenAI), deterministic combinatorial question expansion decoupled from the LLM, and an independent solver using structured Pydantic outputs that verifies every generated answer, while tracking token usage and latency across pipeline stages.

• Co-founded a real-time Agentic AI safety system for construction sites; secured US$100K in Microsoft for Startups Azure credits; Columbia AI Startup Challenge Semifinalist; selected for the UNSW Pre-X Accelerator (25/300 selected); interviewed with Y Combinator and other accelerators (<3% reach YC stage).
• Architected a multimodal vision pipeline using LLaVA and YOLOv11 to perform zero-shot hazard reasoning on live site feeds.
• Containerized and deployed the FastAPI backend to Azure Container Apps using Docker and Azure Container Registry (ACR).

• Shipped a full-stack MVP to 100+ pilot users (UNSW Peter Farrell Cup Top 10 Finalist); engineered a two-stage mentor-matching pipeline using LLM multi-label classification and LLM-as-a-judge A/B testing, improving match quality and user trust.
• Architected Next.js (TypeScript) app with Google Analytics events to track user behavior, designed Airtable data model and automation hooks to recompute matches on specific events.
• Contributed to product discovery, user-journey mapping, and UX workshops to turn insights into features.

A project-based company that collaborates with various clients to deliver AI and automation solutions.

• Built a semantic search workflow over a 10K+ MCQ repository using BERT embeddings, reducing manual drafting workload by 25%, resulting in faster content creation cycles for educators.
• Fine-tuned Seq2Seq T5 on educational Q&A pairs to automate Question Generation, reducing test creation time by ~5 min per question, resulting in quicker quiz/test generation at scale.
• Built a churn prediction model (Logistic Regression, Random Forest) on 3K student/parent survey responses (18.7% churn); achieved ROC-AUC 0.86 (5-fold cross-validation), enabling targeted retention outreach.

A lead generation company aggregating vast amounts of financial and corporate data for decision-making.

• Optimized query performance by migrating 20M records from SQL to Elasticsearch leading to faster searches.
• Engineered an NLP pipeline (spaCy NER, regex parsing) to extract 50K+ structured records from unstructured PDFs.
• Architected 12+ Python scrapers (Scrapy, Selenium) and API integrations to extract data from government & public sources, automating updates for 20K+ records daily and saving 8 hours/week, improving data freshness for decision-making.
• Engineered an internal analytics platform using React/FastAPI to audit data quality, eliminating the marketing team's dependency on engineering support for lifecycle management.

• Built a Flutter image management app (Industry-Academia Partnership) with duplicate detection, geo-tagging, and image enhancements, reducing image search time by 70%.