Pushkaraj Baradkar

Pushkaraj Baradkar

MS Computer Science @ USC Viterbi

Final-year Master's student in Computer Science at the University of Southern California, graduating December 2026.

I build production AI systems: search and recommendation, generative and agentic LLM applications, real-time computer vision, and decision policies learned from logged data.

Seeking full-time Software Engineering, AI/ML Engineering, and Research roles.

View resumeDownload resume
Focus
AI/ML Engineering · Retrieval & RecSys · Agentic Systems
Studying
MS Computer Science, USC Viterbi (Dec 2026)
Now
Research Assistant & CSCI 585 Grader, USC
Based
Los Angeles, CA

News

Scroll
Presented "Quantifying Privacy Risks of Masked Language Models Using Membership Inference Attacks" (EMNLP 2022) in CSCI 699 at USC.
Started as a Grader for CSCI 585 Database Systems at USC, under Prof. Shahram Ghandeharizadeh.
Built a warehouse tote perception pipeline: 0.958 pick score on 120 ARMBench images, 619 ms median on a laptop CPU.
Joined Tabhi as an AI Engineer Intern (Search & Recommendation Systems) in Austin, TX, building hybrid Elasticsearch and vector retrieval over a 450K-item catalog.
Contributed an RPC reliability & failover redesign to the DefiLlama SDK, fixing endpoint override precedence and adding runtime quarantine.
Built AgentPay at the Southern California Blockchain Hackathon: an autonomous multi-agent micro-economy with on-chain payments and staked reputation.
Started as a Research Assistant (Real-Time AI Systems) at USC under Prof. Laurent Itti, building real-time monocular visual odometry and SLAM with a GTSAM factor-graph backend.
Attended WE25, the Society of Women Engineers annual conference, in New Orleans, Louisiana.
Reached the final round of the IndeHub Hackathon 2025, a post-WWDC hybrid hackathon for building on Apple platforms, with Voxel Strides: an iOS app that turns daily tasks into a gamified adventure, with an ARKit companion and an on-device Core ML agent that breaks goals into steps.
Started a Master of Science in Computer Science at the USC Viterbi School of Engineering in Los Angeles.
Graduated with a Bachelor of Engineering in Computer Engineering from the University of Mumbai, GPA 9.13 / 10.
Returned for CodeShastra X, the 10th-anniversary edition of the 24-hour national hackathon at DJ Sanghvi College of Engineering, Mumbai.
Joined YMT Medical as a Machine Learning Intern, building on-device clinical computer vision for micro-lesion detection and deterministic acne grading.
Presented "Blockchain-based Music Streaming Platform using NFTs" at IEEE ICSCSS 2023. Published in IEEE Xplore.
Reached the second round of the Smart India Hackathon (SIH) 2023.
Joined the Rotaract Club of Thane Greenspans as a Community Engagement Volunteer, running coastal cleanups that removed over 100 kg of waste and donation drives for underserved communities.
Won 1st place out of 50+ teams at Syrus 2023, a 24-hour Web3 hackathon organised by CodeCell, VESIT.
Competed at CodeShastra 9.0, a 24-hour offline hackathon organised by the DJ-CSI Student Chapter at DJ Sanghvi College of Engineering, Mumbai.
Began a Bachelor of Engineering in Computer Engineering at the University of Mumbai.
All news →

Service & Activities

  • Community Engagement Volunteer | Rotaract Club of Thane Greenspans
    Jul 2023 – Jun 2024

    Coordinated 10+ volunteers on coastal cleanups and sustainability campaigns that removed over 100 kg of waste, and led logistics for donation drives reaching 30+ people from underserved communities.

  • Mentor | USC Mentorship Programme

    Mentoring incoming graduate students on coursework, research direction, and internship search.

All →

Selected Projects

Machine Learning & Decision Systems

  • HVAC Follow-up Policy Optimization

    Offline RL that learns when not to act. FQI distilled into a depth-10 tree with guardrails; value per thread from −13.46 to +15.25.

Computer Vision & Perception

  • Warehouse Tote Perception

    Where should the suction cup land? FastSAM proposals, a 2-of-3 tote/item vote, and a distance-transform pick point: 0.958 pick score vs 0.31 baseline.

  • Real-Time Visual SLAM with Learned Features

    Learned features survive motion blur where ORB and SIFT do not; a factor graph lets loop closures correct drift retroactively.

Privacy & Trustworthy ML

Agentic AI & LLM Systems

  • Eval-Gated Skill Development

    A skill does not merge unless an eval proves it helps. 17-skill registry, 466-task bank, 3-tier CI gates.

  • AgentPay

    Agents hire and pay each other on-chain. Reputation is bonded capital, so cheating is economically punished rather than merely recorded.

  • InfoDistill

    Raw feeds to readable digest. Zero-shot classification means the topic taxonomy can change without retraining.

Developer Tools & Open Source

  • Godot MCP

    LLM agents control a game engine. The tool boundary is the security boundary, so it has to hold even when the model is wrong.

  • DefiLlama SDK: RPC Reliability & Failover

    User endpoints were silently overridden by defaults. Fixed resolution order, then added quarantine so one bad provider cannot degrade everything.

All projects →

Experience

Scroll
  • AI Engineer Intern, Search & Recommendation Systems | Tabhi
    Austin, TX · May 2026 – Aug 2026

    Architected an end-to-end personalized discovery feed over 450K+ items, and cut cold-feed latency from 8 s to 500 ms.

  • Research Assistant, Real-Time AI Systems | University of Southern California
    Los Angeles, CA · Nov 2025 – Present · under Prof. Laurent Itti

    Monocular SLAM for assistive smart glasses: a C++ GTSAM pose-graph backend cut trajectory error from 29–58 m to 0.09–0.24 m.

  • Grader, CSCI 585 Database Systems | University of Southern California
    Los Angeles, CA · Aug 2026 – Present · under Prof. Shahram Ghandeharizadeh

    Evaluating Database Systems coursework for 200+ students under Prof. Shahram Ghandeharizadeh.

  • Machine Learning Intern, Edge Computer Vision | YMT Medical
    India · Jan 2024 – Aug 2024

    On-device clinical vision for micro-lesion detection (0.83 mAP@0.5) and deterministic acne grading, with ~40% faster edge inference.

  • Software Engineer Intern, Backend & APIs | Technoriya ERP Solution
    India · Oct 2023 – Dec 2023

    Built backend services and APIs supporting scalable data pipelines and cloud-based ML workflows.

Full detail →

Research & Publications

SLAM + ToF paper Under review

with Prof. Laurent Itti · iLab, USC

Part of my research in the iLab at USC.

Blockchain Based Music Streaming Platform Using NFTs figure

Blockchain Based Music Streaming Platform Using NFTs Cited by 7

P. Baradkar et al. · IEEE ICSCSS 2023

A blockchain-based music streaming platform leveraging NFTs for rights management and royalty distribution, enabling transparent and automated compensation for artists.

PaperCodeGoogle Scholar

Paper presentations

Quantifying Privacy Risks of Masked Language Models Using Membership Inference Attacks figure

Quantifying Privacy Risks of Masked Language Models Using Membership Inference Attacks Talk

Mireshghallah, Goyal, Uniyal, Berg-Kirkpatrick & Shokri · EMNLP 2022
Presented in CSCI 699 · Side-Channel Threats in Cloud & LLM Systems · USC · Fall 2026

Paper presentation. Loss-threshold attacks made masked LMs look safe; a likelihood-ratio test against a reference model, scored as an energy gap, lifts attack AUC on ClinicalBERT from 0.66 to 0.90, and to 0.99 per patient.

Write-uparXiv

Skills

Programming Languages

PythonC++JavaJavaScriptTypeScript

AI / ML & Deep Learning

Deep LearningPyTorchTensorFlowTransformersComputer VisionNLPReinforcement LearningOffline RLscikit-learnPandasNumPy

Generative AI & Agents

LLMsRAGAgentic AI SystemsPrompt EngineeringLangChainLangGraphMCPGoogle ADKA2ALangfuse

Retrieval & Recommendation

Elasticsearch BM25Hybrid SearchVector RetrievalFAISSpgvectorTwo-Tower / SASRecOff-policy Evaluation

Computer Vision & Edge

Visual Odometry / SLAMGTSAMOpenCVYOLOv8FastSAM / SAMMediaPipeINT8 QuantizationCoreMLONNX Runtime

Privacy & Trustworthy ML

Differential PrivacydiffprivlibMembership InferenceOff-policy Evaluation

Frameworks & Development

FastAPIFlaskDjangoNode.jsExpress.jsReact.jsNext.jsStreamlitOAuth

Databases & Cloud

PostgreSQLMongoDBSQLRedisSupabaseAWSGCPAzureDockerKubernetesGitCI/CDLinux
All →

Education

Details →