Resume

• Independently designed and built a full-stack estimating and takeoff platform (Python backend, React/TypeScript frontend): document-ingestion pipeline, OCR, spatial registration, and a geometry-aware computation engine with AI-assisted extraction; cut takeoff time from 6 hours to 30 minutes, backed by approximately 900 automated tests
• Led software delivery across Universal Creative and UDX, partnering with stakeholders across 6 business functions to gather requirements and author PRDs, shipping 10 products from concept into production
• Architected 8+ production systems spanning estimating, legal, engineering, safety, finance, and operations — designing shared patterns for service routing, validation, access control, and human-in-the-loop approval
• Led development of a centralized contract lifecycle platform, building ETL and a unified search layer over 10,000+ documents from SharePoint, databases, and enterprise repositories; reduced first-draft contract generation from approximately 2 weeks to under 48 hours
• Built retrieval and evaluation infrastructure over 5,000+ pages of legal documents with semantic search, automated citation validation, and deterministic checks; achieved full citation coverage at an approximately 2% review-error rate
• Delivered a cost-intelligence platform over historical cost data with structured retrieval, query routing, provenance tracking, and a conversational UI; reduced cost-research time 95%
• Built and deployed a production CI debugging agent using Python, FastAPI, GitHub Actions, and LLM tool calling to analyze failed builds, identify likely root causes, and recommend code fixes

• Full-stack software
• Production AI
• Technical leadership

• Built and deployed production video understanding pipelines on GCP using Python, FFmpeg, CLIP/SigLIP, and TimeSformer, generating 100K+ searchable clip embeddings monthly; integrated FAISS/HNSW and BigQuery for semantic retrieval, contributing to 30% higher retrieval accuracy across the multimodal search platform
• Engineered production speech transcription and speaker diarization pipelines using Python, FFmpeg, Whisper, VAD, and speaker-embedding clustering, processing 800+ hours of audio monthly; generated timestamp-aligned, speaker-labeled transcripts and multilingual metadata, contributing to a 40% reduction in manual media search effort across the platform
• Decreased issue resolution time 35% by building an event-driven monitoring service (AWS Bedrock, Lambda, Slack API) for ML pipelines, surfacing drift alerts, exceptions, and health insights in real time
• Reduced ad-hoc reporting effort 40% by building a natural-language-to-SQL assistant (FAISS, Bedrock) serving producers at p95 approximately 1.3s latency
• Built an automated evaluation and regression harness (synthetic queries, oracle sets, precision@k) integrated into the iteration workflow; improved top-5 retrieval by 12 percentage points
• Cut per-request serving cost 28% without quality loss via response caching and lightweight request routing on Databricks
• Engineered and deployed production Apache Airflow DAGs on GCP for 100K+ video segments monthly: FFmpeg preprocessing, Whisper transcription, CLIP/SigLIP embeddings, and BigQuery ETL; parallel execution, automated retries, SLA monitoring, and idempotent backfills achieved 99%+ pipeline success and 50% fewer manual interventions
• Streamlined operations on 2M+ digital assets by automating metadata extraction and ETL (BigQuery, Spark, Cloud Functions); features powered engagement forecasting (LSTM, transformers) that lifted user retention 25%

• AI systems
• Evaluation
• MLOps

Materials Today: Proceedings

Materials Today: Proceedings

CGPA: 3.97

Machine Learning · Deep Learning · Reinforcement Learning · Computer Vision

Helping middle school students design and code robots

Founded and taught the computer science and mathematics curriculum