Hi

I'm

Siddhant Rajhans.

I'm a Machine Learning Engineer working on AI security and LLM evaluation. At a health-tech startup I shipped production LLM and data systems: an agentic AI system that raised engagement about 25%, and streaming pipelines for 1M+ records/month. I publish security research (IEEE, Springer) and contribute fixes to the tools teams use to evaluate, red-team and serve models, at NVIDIA and EleutherAI.

Siddhant Rajhans
25% engagement lift from an
agentic AI system in production
1M+ records / month
through my pipelines
3 peer-reviewed security papers
IEEE · Springer
11 merged open-source PRs
NVIDIA · EleutherAI

01 / Selected Work

Explore My Latest Projects

CortexLab Architecture

CortexLab: Brain Encoding Toolkit

Multimodal fMRI brain encoding toolkit built on Meta's TRIBE v2. GPU voxelwise ridge (Triton), causal modality lesion analysis, brain-alignment benchmarking, 3D brain viewer, live inference from webcam/screen/video. 143 tests, 10 contributors, published on PyPI and HuggingFace.

Agent Evaluation Harness Architecture

Agent Eval Harness for Customer Workflows

A tool-calling support agent in LangGraph with five typed tools (orders, refunds, shipping, inventory, human escalation), and an evaluation harness that scores it on 50 scripted customer scenarios for task success, tool-call accuracy and efficiency. A live run on a local Qwen3 14B resolved 64% with 79% tool-call accuracy.

Bloomberg Sentiment x FinBERT Architecture

Bloomberg Sentiment × FinBERT

Empirical test of whether Bloomberg's NEWS_SENTIMENT_DAILY_AVG can be replicated with open-source FinBERT on a 30-stock S&P 100 universe (2018-2026, 62.8K stock-day obs). Multi-horizon IC, long-short quintile backtest, Fama-French 3 with Newey-West HAC. Replication fails (ρ = −0.26), and that's the point.

DreamStudio Architecture

DreamStudio: AI Cinematic Story Director

AI-powered cinematic story director. Speak naturally, point your camera, and watch scenes materialize as images, video, and music in real-time. Full-stack app with web, mobile, and Python backend.

Data Pipeline Architecture

Messaging-Based Data Pipeline

End-to-end streaming pipeline processing 1M+ records/month with 65s latency (p95). Kappa architecture with event-time processing, watermarking, and fault tolerance built for bursty event ingestion.

Multimodal RAG Architecture

Multimodal RAG for Scientific Literature

Textual + multimodal RAG over 500K+ papers with a LangGraph agent that runs a grounding-validation step, flagging weakly-supported answers instead of returning them (a hallucination guardrail). FAISS retrieval with cross-encoder reranking; 94% recall on an internal evaluation set.

Bike Lane Sentinel Architecture

Bike Lane Sentinel

Computer vision system that monitors illegal vehicle encroachment into NYC bike lanes and automatically alerts NYC DOT. Real-time detection with interactive dashboard and evidence capture.

More on GitHub

02 / Open Source

Open Source & Contributions

Benchmarks, fixes and scorers for the tools teams use to evaluate, red-team and serve models.

PR #3662 · Merged

EleutherAI lm-evaluation-harness

Contributed InfiniteBench, 11 long-context evaluation tasks (retrieval, code, math, novel QA and dialogue, in English and Chinese), to the open-source evaluation framework behind Hugging Face's Open LLM Leaderboard. 799 lines across 15 files, with scoring matched exactly to the official implementation.

3 PRs · Merged

NVIDIA NeMo Agent Toolkit

Fixes to its red-teaming and evaluation packages: a judge failure on one item no longer throws away a whole red-team run, a target that matches an object, list or null no longer counts as an attack on an emptied input, and the trajectory evaluator reports on its 0 to 1 scale.

7 PRs · Merged

NVIDIA inference and training stack

TensorRT-LLM: raw logprobs now come from the logits saved before biases and penalties rewrite them. TransformerEngine: MXFP8 and NVFP4 tensor ops, and misaligned grouped-GEMM operands. Megatron-Bridge: FP8 checkpoint import, RoPE on export, vocabulary padding. Dynamo: named chat templates.

4 PRs · In review

Microsoft PyRIT

AI red-teaming scorers for Microsoft's PyRIT: CyberSecEval prompt injection, HarmBench and StrongREJECT, plus an approved fix so the fuzzer counts each jailbroken response once.

03 / Experience

Built Production Systems & Shipped Real Products.

Oct 2023 – Dec 2024

Junior Machine Learning Engineer

Seed-Stage Health-Tech Startup (NDA)

  • Led the retrieval and prompt-orchestration workstream of an agentic AI system powering personalized clinical-facing interactions, raising user engagement by about 25%.
  • Built transformer-based NLP and multimodal retrieval for clinical notes, lab reports, and diagnostic documents with clinical and product teams, improving retrieval accuracy by 30%.
  • Lowered the agent's serving cost by routing simple requests to a small fine-tuned model and caching retrieval, keeping the large model for the hardest 25% at under 5 s p95.
  • Designed and shipped streaming ML pipelines on AWS (Kafka, Airflow, Docker) with the data engineering team, processing 1M+ healthcare records/month at under 65 s p95 latency.
  • Built monitoring and validation checks for data drift, prediction quality, and data integrity across clinical ML workflows.
AWSKafkaAirflowDockerNLPAgentic AI
Apr – Sep 2023

Software Development Engineer Intern

Seed-Stage Health-Tech Startup (NDA)

  • Reduced average API request latency by 40% by moving synchronous flows to async processing, and added caching and request batching to ML model serving.
  • Strengthened API security for sensitive healthcare data with senior engineers: input validation, authentication middleware, and audit logging.
ML ServingAPIsBackend OptimizationSecurity

04 / Toolkit

Technologies and Tools I Work With

LLM Evaluation & AI Security

LLM Evaluationlm-evaluation-harnessRed-teamingPyRITNeMo Agent ToolkitPrompt-injection & Jailbreak ScoringGuardrailsGrounding / FaithfulnessModel Monitoring & DriftResponsible AI (NIST AI RMF)

LLM, Agents & RAG

LangGraphRAGAgentic OrchestrationTool CallingPrompt EngineeringBenchmark Design

Core ML & Deep Learning

PyTorchTensorFlowScikit-learnTransformersHugging FaceCNNs

Data, Infra & MLOps

PythonSQLApache SparkKafkaFAISSPostgres/PgvectorAWS (SageMaker, S3)DockerAirflowMLflowW&B

05 / Research

Research Publications

Peer-reviewed work in AI security and trust: deepfake detection, ML for cybersecurity, and threats in AI-integrated cloud systems.

06 / Writing

From the Blog

Notes on building ML systems, evaluating frontier models, and lessons from production.

Read all posts

07 / Education

Education & Credentials

2025–2026

MS in Machine Learning

Stevens Institute of Technology

Hoboken, NJ
2019–2023

B.Tech in Computer Science & Engineering

Swami Rama Himalayan University

Dehradun, India

Certifications

Weights & Biases Training & Fine-tuning LLMs Verify credential
Stanford / Coursera Machine Learning Verify credential
NVIDIA Generative AI with Diffusion Models Verify credential
IBM Cloud Computing Diploma Verify credential

08 / Contact

Let's build something.

Graduating December 2026 and starting January 2027, with three years of US work authorization (STEM OPT). Open to ML engineering, applied AI and research-engineering roles, especially in AI security and LLM evaluation. Based in Hoboken, NJ (New York area); open to relocating anywhere in the US.