Back to home

Engineering Portfolio & Master CV

Saksham Kapoor

sakshamk@umd.edu·LinkedIn·GitHub
Download PDF

Executive Profile

High-impact Machine Learning Researcher and Systems Engineer specializing in ML compilers, distributed streaming runtimes, and population-scale agentic evaluation. Appointed as Workshop Area Chair (NeurIPS, EMNLP), Program Committee (AIES), federal proposal reviewer (NAIRR Pilot), and White House AI Challenge judge.

🎓 Education & Fellowships

University of Maryland, College ParkExpected May 2028

B.S. (Undergraduate)

AI/ML Fellow (Break Through Tech AI at Cornell)Handshake MOVE Fellow (Model Validation Expert)AWS ScholarStartup Shell Fellow (UMD incubator)

🛠 Technical Skills

Languages & Frameworks

PythonGoC#JavaTypeScriptPyTorch.NET 8Spring BootReact NativeNode.jsLangGraph

Systems & Infrastructure

Kubernetes (GKE)DockerAWS (SQS, DynamoDB, Bedrock, S3)RedisPostgreSQLRaft ConsensusClickHouse

AI & Interpretability

TransformerLensWhisperXSAM 2Transformers (RoBERTa, BERT)OpenCVGANsLLM Fine-tuning (GRPO, LoRA)

GIS & Visualization

GeoPandasGDALRasterioMesa (Agent-Based)React Three Fiber (R3F)Three.jsMapbox GL JSLaTeX

💼 Professional Experience

International Student Research Mentorship (ISRM)

AI Research Mentor

Jul 2026 - Present
  • Mentor a cohort of 7 students through independent AI/ML research, from problem formulation and literature review to experimental design and manuscript planning.7mentees
  • Guide reproducible evaluation, responsible AI research practices, and scientific writing for students preparing original work for peer-reviewed submission.

Trufflow

Machine Learning Engineer Fellow

Sep 2025 – Dec 2025
  • Real-Time Monitoring: Built an enterprise anomaly detection system handling 1.2M+ events/sec peak load utilizing Polars data pipelines.1.2M+events/sec
  • Precision Optimization: Designed a graph-aware model mapping component topology, achieving a 73× PR-AUC improvement over static baselines.73×PR-AUC improvement
  • Debugging Workflows: Replaced static thresholds with relative deviation scaling, resulting in a 28× debugging speedup for unpredictably scaling tenants.28×debugging speedup

Handshake AI

Model Validation Expert (MOVE) Fellow

Jul 2025 - Nov 2025
  • Applied academic subject-matter expertise and human judgment to evaluate, challenge, and improve frontier LLM outputs through specialized validation workflows.
  • Reviewed model behavior for reasoning quality, domain specificity, and safety-relevant failure modes, translating qualitative feedback into actionable model-evaluation signals.

Child Development Lab, UMD

Machine Learning Researcher (CV & NLP)

Jan 2025 – Present
  • NIMH-Supported Study Infrastructure: Optimized SAM 2 video processing for the Mother-Child Dynamics study; logged propagation time of 26m 21s across 5,396 frames, compared with an estimated ~112-minute baseline.26m 21slogged propagation
  • FFmpeg Speech Pipeline: Coordinated end-to-end speaker diarization integrating WhisperX and pyannote.audio for 100+ dyadic sessions in the NIMH-supported Mother-Child Dynamics study.100+sessions
  • Sentiment Contagion: Fine-tuned RoBERTa models to analyze emotional synchronization using Linear Mixed Effects Models in R, validating bidirectional associations across positive (β = .27) and negative (β = .30) valences.β = .27positive valenceβ = .30negative valence

Viewing mode

Top experience & flagship projects per vertical

Targeting deep-stack engineering problems inside Tier-1 open-source ML compilers, edge execution engines, and cloud orchestration systems — verified and merged by recognized industry maintainers.

Toolkit:
Google XLAPyTorch Core & InductorCore ML / MILExecuTorch XNNPACKApache AirflowOpenSSLNetflix MaestroSAM 3C++CUDATritonPython

High-impact core infrastructure PRs merged upstream into leading ML compilers, edge execution runtimes, and distributed workflow orchestrators.

+1 more project — switch to Full CV to view all

Engineering robust agentic architectures, self-healing execution loops, and high-throughput observability pipelines for complex LLM deployments.

Toolkit:
LangGraphGemini 2.5 FlashClickHouseGo (Golang)Playwrightaxe-coreDockerKubernetes

An agentic-web accessibility engineer that audits live websites, traces WCAG violations to source components, proposes self-healing patches, and packages fixes into pull-request-ready changes.

+2 more projects — switch to Full CV to view all

Engineering high-fidelity, low-latency audio processing pipelines — from contact-center telephony and live voice proxying to academic multimodal sentiment extraction.

Toolkit:
Go (Golang)WebSocketsWhisperXpyannote.audioTwilio SIP/RTPG.711 PCMUGemini 1.5 Flash Native APIDockerGCP Cloud Run

Bypasses legacy "Sandwich" pipeline latencies (Whisper STT → LLM → TTS) by proxying real-time audio streams directly to Gemini 1.5 Flash's Native Multimodal Audio API.

+2 more projects — switch to Full CV to view all

Independently selected to validate, review, and evaluate cutting-edge Machine Learning research and implementations for presidential initiatives, top-tier international conferences, and cloud provider symposia.

  • National AI Research Resource (NAIRR) Pilot (2026): Appointed to the review panel to assess AI research proposals and computing-resource allocation requests.panelistNAIRR Pilot
  • State & Regional Judge - White House AI Presidential Challenge (2026): Evaluated advanced student AI applications and technical architectures for the US President-backed K-12 artificial intelligence literacy and development initiative.
  • Technical Judge — 10+ National AI Hackathons (2024–Present): Served as an expert panelist auditing software architecture, model parameters, API pipelines, and algorithmic efficiency at Anthropic × Maryland, HackGT, HackTX, TartanHacks, Bitcamp, HackDuke, and Hacklytics.
  • Judge - COVID Information Commons Student Paper Challenge (2025): Evaluated student COVID-19 research papers connected to the NSF-funded CIC research ecosystem, applying criteria around intellectual merit, broader impact, reproducibility, and public-health relevance.CIC Student Paper Challenge

+2 more projects — switch to Full CV to view all

Press ⌘+K to search