Siddharth Raj
Open to Research & Full-time ML/AI Roles

Hi, I'm

Siddharth Raj

ML/AI Researcher  ·  MS CS @ University of Chicago  ·  Ex-Walmart Global Tech

I build ML systems that are both theoretically grounded and production-ready. My work spans RLHF alignment pipelines, geometric computer vision with transformers, and high-throughput NLP systems.

3.8/4.0
MS GPA · UChicago
3+ yrs
Industry ML/SWE
7+
ML Research Projects
Top 5%
Codeforces Globally
12th
ACM ICPC NA Regionals
say hi
01About

Who I am

I'm a Master's student in CS (AI Specialization) at the University of Chicago, with 3+ years at Walmart Global Tech building production LLM systems and distributed backends.

I work on computer vision, LLM alignment, and efficient ML systems — and I care most about turning research insight into things that run in the real world.

B.Tech from IIT BHU. Candidate Master on Codeforces.

Education

University of Chicago

MS Computer Science — AI Specialization

GPA 3.8 / 4.0 · Sep 2025 – Dec 2026

Indian Institute of Technology, BHU

B.Tech

GPA 8.7 / 10.0 · Jun 2018 – May 2022
02Research & projects

What I've built

01 Vision-Language · Grounding · LoRA Spring 2026

GRAFT — Region-Grounded Fine-tuning of SigLIP

Taught spatial grounding into a frozen SigLIP-B/16-384 with a LoRA adapter (295K params, 0.14%) and a FILIP-style region loss on phrase→bbox pairs. The dominant gain came not from the loss or capacity but from eliminating train/eval representation mismatches (MaskCLIP bypass + EOS alignment) — raising Flickr30k Pointing Game from 14.74% to 20.76 ± 0.41% (3-seed, 6× lower variance) and 26.16% test PG at 5 epochs. Florence-2 auto-labels reach 23.35% with zero human boxes.

  • +11.4pp Pointing Game
  • 295K params (0.14%)
  • 6× lower variance
  • Florence-2 auto-labels

PyTorch · SigLIP · LoRA / PEFT · FILIP · Flickr30k · W&B

02 LLM Alignment · Reinforcement Learning Jan – Feb 2026

RLHF Fine-tuning Pipeline for Llama-2-7B

End-to-end 3-stage alignment pipeline: SFT → Reward Model (Bradley-Terry pairwise ranking, 82% human agreement) → PPO optimization with KL-divergence penalty to prevent reward hacking. LoRA adapters cut trainable params from 6.7B to 33M (0.5%), enabling single-GPU training. W&B tracking throughout.

  • +35% safety score
  • −70% GPU memory
  • 82% human agreement
  • +20% output diversity

PyTorch · HuggingFace TRL · LoRA / PEFT · PPO · Llama-2-7B · W&B

03 Computer Vision · Transformers · Geometry Jan – Mar 2026

Epipolar Geometry in MatchFormer

Injected differentiable epipolar constraints into a transformer-based feature matcher via a soft fundamental-matrix mask on the confidence matrix. Fine-tuned with focal loss + LoRA adapters on ScanNet; diagnosed confidence distribution shift post-training — threshold recalibration to 0.10 recovered 436 matches/pair while preserving geometric precision.

  • −94% reprojection error
  • +18% Precision@3px
  • 546× over baseline

PyTorch · LoRA · ScanNet · PyTorch Lightning · Fundamental Matrix · Focal Loss

04 Game Theory · Data Science · NYC Data Spring 2026

NashCab — Nash Equilibrium in Ride-Sharing

Modeled Uber/Lyft as a Mean-Field Game over 259 NYC TLC zones using real HVFHV trip data. Backward induction + fictitious play finds the Nash equilibrium in 166 iterations — drivers at JFK queue for high-fare Manhattan runs. Interactive map shows driver redistribution and hourly demand across the city.

  • PoA = 1.52
  • 589K rides
  • 166 iters
  • 10,979 drivers

Python · D3.js · Mean-Field Game · Backward Induction · NYC TLC Data · Fictitious Play

05 Computer Vision · MLOps · Deployment Sep – Dec 2025

Image Classification MLOps Pipeline

Custom PyTorch CNN on CIFAR-10 (BatchNorm + residual connections, Adam + StepLR). Exported to ONNX Runtime via FastAPI for production inference; deployed on AWS ECS via ECR. Grad-CAM explainability surfaced systematic animal misclassifications — targeted augmentation boosted accuracy 2.5%. Model and weights published to HuggingFace.

  • 93% accuracy
  • −40% inference latency
  • HuggingFace published

PyTorch · ONNX · FastAPI · Grad-CAM · AWS ECS · Docker

06 NLP · Topic Modeling · Analytics Sep – Dec 2025

NLP Customer Feedback Analyzer

4-stage pipeline over 10,000+ reviews at ~9,100 reviews/sec. Combines VADER lexical sentiment (69.3% star-rating agreement) with BERTopic — sentence transformers + UMAP dimensionality reduction + HDBSCAN clustering — to auto-discover 44 latent topics and extract 10 composite priority pain points. Served via a Flask analytics dashboard.

  • 9,100 reviews/sec
  • 44 auto-clusters
  • 69.3% agreement

Python · VADER · BERTopic · UMAP · HDBSCAN · Flask

07 Systems · Parallel Computing · Graphics Nov – Dec 2025

Parallel Ray Tracing Simulator

Multi-threaded ray tracer in Go implementing global illumination, reflections, and shadows. Work-stealing scheduler with lock-free deques distributes tile partitions across 15 cores for load balancing under variable render costs. Profiled on UChicago Peanut cluster with Go's pprof — 12× speedup over sequential baseline.

  • 12× speedup
  • 15-core scaling
  • Lock-free deque

Go · Work-Stealing · Lock-free · pprof · Concurrency

08 Computational Math · Computer Graphics 2025

Procedural Physics Simulation & Mesh Parameterization

Built LSCM UV parameterization from scratch using sparse Cauchy-Riemann equations over bmesh topology — mathematically optimal angle-preserving mapping for curved surfaces. Modeled multi-body collision dynamics via quadratic B-Splines with C0 knot multiplicity at impact points. All geometry, shader nodes, and particle systems generated programmatically via the bpy API.

  • LSCM from scratch
  • B-Spline physics
  • 100% procedural

Python · Blender (bpy) · LSCM · B-Splines · Procedural Shaders · Sparse Matrices

03Experience

Where I've worked

Jun 2026 – Sep 2026
New York, US

Amera Health Solutions

Machine Learning Intern
  • Fine-tuned vision-language models (VLMs) to improve OCR accuracy on noisy, low-quality scanned medical documents, benchmarking candidate checkpoints against the production text-extraction pipeline to guide model selection and error analysis
  • Built a synthetic data generation pipeline producing anonymized, PHI-free samples, with clustering to auto-detect new document templates and add them to the generation set — expanding coverage of rare types without exposing real patient data
  • Developed an evaluation harness (accuracy dashboard) tracking per-field and per-document-type error rates across the OCR → parsing → JSON-extraction pipeline to guide fine-tuning and catch quality regressions before release
Jun 2026 – Sep 2026
Remote

Google Summer of Code

Open Source Contributor · Catrobat
  • Distilled a Qwen3-VL 30B teacher into a 3B student, matching its captioning quality at far lower inference cost
  • Trained an octopus-detection classifier on CLIP embeddings with an MLP head to filter relevant octopus footage frame-by-frame
  • Combined the classifier with motion-detection filtering to distill 1TB of raw footage into ~50GB of clean, caption-ready clips
Oct 2025 – Present
Chicago, IL

University of Chicago

Software Engineer — Computing Services
  • Shipped a speech-to-text research platform on AWS Amplify (Next.js, Python) enabling researchers to train and deploy custom language models; NIST 800-171 compliance; GraphQL APIs over S3 + DynamoDB; Okta / Grouper / Cognito SSO
  • Built multi-level auth and authorization workflows via AWS Cognito + Okta securing 200+ university staff across role-based permission tiers (Inventory Management Platform, DynamoDB)
  • Built Python middleware syncing JIRA with SolarWinds via webhooks and CrowdStrike via APIs, automating IT operations event routing
Jun 2022 – Aug 2025
Bengaluru, India

Walmart Global Tech

Software Engineer III
  • Developed an internal AI chatbot using GPT-4o, LangGraph, and Milvus (RAG) — reduced seller support tickets by 30%; Python backends + React UI, end-to-end
  • Built an Elasticsearch Translation Memory System using KNN + fuzzy search — cut translation costs 30% and enabled 80% localization coverage across seller apps
  • Designed a Redis caching layer over MongoDB — dropped P95 API latency from 800ms to under 200ms across 10+ microservices
  • Parallelized seller onboarding Java workflows — 36× faster (3 days → 2 hours)
  • Designed 10+ Spring Boot microservices with Kafka pipelines (DLQs, zero message loss); mentored 3 junior engineers on production debugging and RAG design
  • Presented "WICR: Smarter and Contextual Language Translation Reviews" at Walmart's internal research summit — top 25 of 800 submissions
04Skills

Technical stack

ML / DL Frameworks
PyTorchTensorFlowHuggingFace TRLPEFT / LoRATransformersScikit-learnXGBoostPySparkGrad-CAMONNX Runtime
LLMs & AI Systems
LangChainLangGraphRAGGPT-4oLlama-2MilvusPineconeWeights & Biases
Languages
PythonC++GoJavaJavaScriptC
Cloud & Infrastructure
AWSGCPAzureDockerKubernetesKafkaElasticsearchRedis
Databases
MongoDBPostgreSQLDynamoDBMilvusAzure Cosmos DBCassandraDB
Research Methods
RLHF / PPOFine-tuningEpipolar GeometryTopic ModelingExplainability (XAI)Parallel Computing
05Recognition

Achievements

Candidate Master — Codeforces

Top 5% of competitive programmers globally. Strong foundations in algorithms, combinatorics, graph theory, and mathematical problem-solving that inform how I approach ML research problems.

Ranked 12th — ACM ICPC North America Regionals

Top 12 finish among hundreds of teams at one of the most prestigious collegiate programming competitions in North America.

Research Paper — Walmart Global Tech Innovation Summit

"WICR: Smarter and Contextual Language Translation Reviews" — selected among top 25 of 800 submissions for innovation and real-world impact on localization at scale.

06Contact

Let's talk.

I'm open to research collaborations, full-time ML/AI roles, and interesting conversations about alignment, geometric vision, and building systems that actually work.