Machine Learning Engineer · TikTok

I build recommendation systems that decide what millions of people see next.

Recall, ranking, and conversion modeling for TikTok Local Services. Before that: LLM agents at Scale AI, volatility forecasting at NYU's V-Lab, and an M.S. in Computer Engineering from NYU.

Ask my AI anythingSee my work
recall → ranking → re-rank · production recsysM.S. CompE · NYU ’26prev · Scale AI, NYU V-Lab

00 · Ask my AI

Skip the scrolling, just ask.

A small chat that answers questions about me, powered by a quantized embedding model running in your browser via transformers.js. No server, no API keys, and your questions stay on your device.

ask-moulik · local RAG

MiniLM-L6-v2 · loads on first question

Ask anything about me. Your question gets embedded and matched against a small knowledge base, all inside your browser. Nothing gets sent anywhere.

first question downloads the model (23MB, one time)

01 · Experience

Where I’ve worked

  1. Machine Learning Engineer · TikToknow

    June 2026 — Present

    Local Services · US Transaction Team · San Jose, CA

    • Work across the full recommendation stack (recall, pre-ranking, ranking, re-ranking) for US Local Services (accommodations, travel, food, leisure, beauty).
    • Build and iterate on CVR models, engineering signals from user sessions captured at exposure time along with click-through and dwell-time behavior.
    • Own the model lifecycle end to end: spinning up training instances, running feature backfills, and shipping scoring algorithms to production.
    • Run offline evaluations and online A/B experiments, tying model wins directly to orders and GMV growth.
    RecommendationRankingCVRA/B Testing

  2. Generative AI Engineer Intern · Scale AI

    Sep 2025 — Dec 2025

    Generative AI · San Francisco, CA

    • Improved LLM reasoning on math and programming tasks through structured prompting, debugging model-generated Python/C++, and iterative feedback loops.
    • Built agentic workflows with MCP-style tool calling: models that retrieve context, validate code against test harnesses, and self-correct via action-observation loops.
    LLMsAgentsMCPEvals

  3. Machine Learning Research Assistant · NYU — Volatility Laboratory (V-Lab)

    Jan 2025 — Dec 2025

    Stern Volatility Lab · New York, NY

    • Rebuilt V-Lab's site search into an AI-powered system: multiple agents handling query understanding, rewriting and expansion, retrieval, and re-ranking, so researchers actually find the right model pages.
    • Built an LLM-powered sentiment trading system on FinBERT and financial news, improving risk-signal accuracy by 15% alongside SRISK and GARCH market forecasts.
    • Designed a scalable backtesting framework and real-time volatility pipelines with GARCH-LSTM and Elasticsearch, cutting data retrieval time by 25%.
    SearchLLM AgentsFinBERTGARCH-LSTMElasticsearch

  4. Data Engineering Intern · Cornspring

    Jun 2025 — Aug 2025

    Software Engineering & DevOps · New York, NY

    • Built cloud ETL pipelines on AWS and Azure ingesting tens of millions of equity-price records daily; engineered features (moving averages, RSI, volatility) that lifted model performance by 60%.
    • Shipped CloudWatch/EventBridge alerting, Terraform infrastructure-as-code, and CI/CD that cut deployment cycles from hours to minutes.
    AWSTerraformETLSageMaker

  5. Data Scientist · MIKO

    Jul 2023 — Aug 2024

    AI Research & Development · Mumbai, India

    • Engineered a reinforcement-learning recommendation engine for skill-based games in PyTorch, validated with A/B testing, which drove a 25% lift in user retention.
    • Led multilingual NLP pipelines with embeddings and NER for speech recognition (+30% accuracy) and improved open-domain QA with DPR + RAG (−15% unanswered queries).
    • Integrated LLMs with anomaly detection to flag suspicious interactions and protect platform integrity.
    PyTorchRecSysRAGNLP

Earlier

  1. Quantitative Research Intern · Sykes & Ray Equities

    Oct 2022 — Dec 2022

    • Developed and backtested trading models with Monte Carlo simulations and alpha hedging, using graph theory to spot patterns, good for a 12% boost in portfolio performance.
    • Automated options strategies like straddles and iron condors in Pandas over NSE data (SQL + APIs), reaching 85% model accuracy through feature engineering and risk assessment.

  2. Machine Learning Intern · Bipolar Factory

    Jan 2022 — Mar 2022

    • Built end-to-end deep learning projects with CNNs in TensorFlow and deployed them on AWS for real-time use.
    • Trained YOLOv4 and SSD object detection models (89% accuracy) and served them through Flask APIs, working with the frontend team on the React/Node integration.

  3. Deep Learning Engineer Intern · Myraa Technologies

    Jan 2022 — Feb 2022

    • Fine-tuned Random Forest and CNN models for HR analytics, sharpening data-driven hiring decisions.
    • Built a TensorFlow binary classifier that detects disaster tweets at 91% accuracy, with a React UI for visualizing results.

  4. Software Engineering Intern · Ares Data

    Dec 2021 — Jan 2022

    • Designed scalable microservices with Spring Boot and Node.js, adding Redis caching and load balancing for 25% faster data retrieval on MongoDB-backed services.
    • Set up an automated test suite (JUnit, Mocha) that raised coverage by 30%, plus Jenkins CI/CD pipelines for faster, consistent deployments.

  5. Corporate Relations Strategist · AIESEC

    Feb 2020 — Jan 2021

    • Ran B2B lead generation and client relations for the Incoming Global Talent program, working with MNCs and SMEs on internship placements.

02 · Projects

Things I’ve built

Flagship · 2025

Distributed Biometric Verification Pipeline

Production-grade face-verification system: ArcFace/iResNet-50 trained with DDP + AMP, tracked in MLflow, and served as quantized ONNX on Triton Inference Server with OpenVINO for high-throughput, low-latency inference. Canary rollouts and drift-triggered retraining keep it reliable in production.

PyTorchTritonONNXMLflowKubernetesTerraform

2024

Cross-Domain Soft Prompt Tuning

Cross-domain adaptation for low-parameter LLMs via prefix tuning on T5-large: 90%+ sentiment accuracy across domains while training under 1% of parameters, with LoRA for efficient few-shot fine-tuning.

LLMsLoRAPrompt TuningT5

2024

LLM-Powered Financial Sentiment Trader

Real-time NER + sentiment on financial news with DistilBERT/FinBERT over Point72's CSP streaming library. The sentiment strategy beat traditional baselines (Sharpe 0.84) on a 20GB global news dataset.

FinBERTStreamingTradingNER

2023

Smart Interactive Marketing

CNN fashion-product classifier on DeepFashion (800k images, 88% accuracy) wired into a React app for real-time recommendations, +30% product engagement.

CNNTensorFlowReact

2023

Speech Emotion Recognition

CNN speech-emotion classifier (82% accuracy) built on MFCC features and Wav2Vec2, served in real time through Gradio and a Flask API.

Wav2Vec2Signal ProcessingGradio

2022

Smart Mart

Smart shopping basket: YOLOv3 product identification connected to an e-commerce platform, plus a behavior-clustering recommendation system.

YOLOFlaskRecSys

03 · Education

Where I studied

New York University

2024 — 2026

M.S. Computer Engineering

Machine Learning · NLP · Deep Learning · MLOps · Computer Vision · Web Search Engines

University of Mumbai

2019 — 2023

B.Tech, Electronics & Telecommunication Engineering

Data Structures & Algorithms · Neural Networks · Big Data Analytics · Image Processing

Achievements

  • Finalist — Indian Institute of Project Technology — Led an autonomous healthcare-delivery drone system (Python + OpenCV) to the national finals.
  • Winner — The Math Company Triathon — Won the competitive analytics assessment outright, earning a direct job offer.

04 · Skills

What I work with

Languages

PythonC++CJavaSQLRJavaScript

ML / AI

PyTorchTensorFlowScikit-learnHugging FaceLangChainOpenCVMLflowTritonONNX

Data & Infra

SparkHadoopAirflowElasticsearchSnowflakeRedshiftDockerKubernetesTerraformAWSGCPAzure

Web & Backend

ReactNext.jsNode.jsFlaskDjangoSpring Boot

05 · Contact

Let’s talk recsys, LLMs, or ML infra.

I’m building at TikTok these days, but I’m always up for a good conversation. Email is the fastest way to reach me.

mps10088@nyu.eduLinkedIn ↗GitHub ↗Resume ↗