Most days start here — laptop open, systems taking shape.
NOWComputer vision research · FAMS Lab, NYU Tandon
RECENTLYBuilt Prism · multi-provider LLM gateway
01Selected Work
Case Study / 001AI Infrastructure · Jul 2026
Prism
Teams that scale past calling a provider directly from the app run into unpredictable cost, no visibility into failures, PII/prompt-injection exposure, and no resilience when a provider degrades — usually all at once.
245–259 req/scombined throughput in a local load test (one machine, mixed Ollama + mock-provider traffic) · 0% unintended errors
0.00%fast-model failure rate with local Ollama made unreachable — traffic failed over to mock providers; 1 stray timeout in 31,488 requests
Fully local document Q&A: ingests PDFs, Word files, spreadsheets, and images (with OCR), retrieves with hybrid FAISS semantic and TF-IDF keyword search, reranks with a cross-encoder, and answers through a local Ollama model, with source citations on every answer.
~17 ms medianhybrid retrieval + reranking over a generated 1,200-file / 4,750-chunk corpus — local benchmark on one machine; the OCR path wasn't part of it
~0.4 s medianend-to-end answer including local llama3.2 generation, same local benchmark
Event discovery and ticket booking with 10-minute Redis inventory holds, server-side pricing the client can't override, and checkout, webhooks, and refunds integrated against Hyperswitch's sandbox — test payments only, not live commercial processing.
15 → 10concurrent requests vs. tickets left in the spike test — exactly 10 orders succeed, no oversell
10 mininventory hold per order, expired lazily with payment-intent cancellation first
Data & AnalysisAug 2026 · Built for a Teiko take-home
Immune Cell Population Analysis
A reproducible pipeline and dashboard for the cell-count dataset supplied with the assignment (framed as a clinical-trial scenario): a normalized SQLite schema, per-sample relative frequencies across five immune-cell populations, and responder vs. non-responder statistics reported with effect sizes.
15 testspytest suite run by GitHub Actions CI on every push and pull request
0 of 5populations significant after FDR correction — reported as a null result, with effect sizes
Real-time flight-congestion pipeline: Kafka ingestion and Spark Structured Streaming risk scoring over live OpenSky data and a ~475,000-record generated historical dataset, with InfluxDB, MongoDB, and a Streamlit dashboard.
Streamlit tool that runs Slither static analysis on Solidity contracts, ranks findings by severity, and summarizes each in plain language with a local FLAN-T5 model; a second tab forecasts next-block Ethereum base fees with a small PyTorch GRU.
Serverless photo search on AWS: S3 uploads trigger a Lambda that labels images with Rekognition and indexes them in OpenSearch; natural-language queries are parsed with Lex and served through API Gateway, with infrastructure in CloudFormation.
Serverless dining-concierge chatbot on AWS: Lex for conversation, Lambda for business logic, and restaurant data scraped from Yelp into DynamoDB and Elasticsearch, served through API Gateway.
Ongoing academic research at the Flexible AI-Enabled Mechatronic Systems Lab: developing a real-time pipeline that detects and localizes kidney stones in microscopic-camera video, covering image/video preprocessing, model development and evaluation, and live-stream inference.
Earlier lab work on endoscopic image analysis: YOLOv5 object detection, U-Net segmentation, and visualization in 3D Slicer.
Involved through NYU's Vertically Integrated Projects program since Fall 2025; Volunteer Research Assistant since September 2026.
Developed an Azure-connected RAG application with LangChain and FAISS, letting 50+ internal staff search 1,000+ enterprise documents and reducing information retrieval time by approximately 70%.
Automated document ingestion, embedding generation, and semantic indexing to keep the knowledge base current.
Built Power BI Q&A semantic models supporting natural-language queries and self-service analytics.
RAGLangChainFAISSAzurePower BIPython
Artificial Intelligence Researcher (Part-time)
University of Essex
Sep 2023 – Aug 2024
Remote, UK
Transformer-based NLP research on early detection of body dysmorphic disorder: text classification, named entity recognition, and topic analysis over ~10,000 annotated samples.
Findings were published in Human Behavior and Emerging Technologies (2026).
NLPTransformersText ClassificationNERResearch
Global Technical Support Engineer
Ciena
Jan 2024 – Jun 2024
Gurugram, India
Troubleshot optical and packet-network issues for global enterprise and telecom customers using logs, packet captures, and CLI diagnostics, handling 50+ escalations a month across 15+ clients.
Wrote automation scripts and technical documentation that cut case resolution time by approximately 25%.
Co-developed a pilot-scale serious game integrating CBT techniques with AI-driven analysis to support body image distress and appearance-related concerns. The AI model interpreting user input reached 97.15% accuracy on held-out test data, outperforming a BERT-based baseline, with preliminary results pointing to gamified digital interventions as a scalable, accessible complement to traditional CBT.
97.15%
AI model accuracy interpreting user input on held-out test data
Introduced a CNN-based model for American Sign Language recognition trained on color and skeleton-plotted image datasets, examining how preprocessing strategy affects recognition across varied backgrounds.
Co-authored chapter investigating the collaborative integration of Blockchain, the Metaverse, and IoT toward decentralized smart urban ecosystems. Explores resource optimization, citizen experience, security, and sustainability — with focus on data privacy via blockchain and real-time urban service management via IoT.
Case Study on Ethical AI-Based Decision-Making in E-Commerce Industrial Sector: Insights on McDonald's and Deliveroo
Wiley — Ethical Decision-Making Using Artificial Intelligence: Challenges, Solutions and Applications (Chapter 12)
Anushka Singh · Naman Tyagi · Dolly Sharma
Explored real-world implementation of AutoML principles at McDonald's, which optimizes food menus based on user preferences, and Deliveroo, which improved delivery times and service quality through AutoML. Examined AutoML's impact on business workflows alongside the ethical considerations organizations should weigh when adopting it.
I'm a software engineer working across applied AI, LLM systems, computer vision, and cloud — from Django and FastAPI services and RAG applications to real-time computer vision research. I completed my MS in Computer Science at NYU Tandon in May 2026.
Brooklyn, NY
Brooklyn, NY · London · New Delhi
Education
Master of Science — Computer Science
NYU Tandon School of Engineering
Aug 2024 – May 2026
B.Tech — Computer Science & Engineering (Hons. in AI & ML)
Amity University Noida
Jul 2020 – Jun 2024
Certificates & training
Certificate: Google Cybersecurity Certificate
Courses: Google Cloud Big Data and Machine Learning Fundamentals · Introduction to Digital Transformation with Google Cloud · Crash Course on Python (Google) · Optical Communications — Associate (Ciena Technology Program)
Away from the keyboard
Badminton and tennis, chess, always hunting for a new restaurant, and planning the next trip.
