Sidhartha Reddy Potu
I hold an MS in Computer Science from NYU with a strong background in ML, Computer Vision, and Software Engineering. I am currently a Partner Engineer, Generative AI at Meta. Most recently, I won first place at the OpenEnv Hackathon in SF, where my team built an RL environment that teaches a small LLM to diagnose and fix real Kubernetes incidents from scratch.
Experience
Partner Engineer, Generative AI
Meta · Menlo Park, CA, USA
June 2026 - Present
Applied Reinforcement Learning Engineer
Centific · Remote
April 2026 - June 2026
- Worked independently with one teammate to develop Kube SRE Gym into a research paper
- Contributed to an early-stage RL environments project for enterprise agentic workflows
SWE Intern
Granica AI · Mountain View, CA, USA
September 2025 - March 2026
- Built a dual-lane FIFO load balancer for a 400+ node K8s cluster, doubling throughput (10 to 20+ req/worker)
- Eliminated runaway K8s API calls at 20k concurrency by caching scaler state in Go
- Resolved lock contention and races in load-balancer goroutines, improving CPU utilization from 87% to 97%
- Designed local scale-validation and integration test suites for the load-balancer and worker routing logic
- Built a benchmark suite to validate performance and correctness as a release gate for a new product line
MTS Intern
Neopolis AI · Mountain View, CA, USA
August 2025 – September 2025
- Refined LiveKit-based LLM voice AI Agent prompts to improve instruction-following and downstream tool inputs
- Used Langfuse observability to inspect agent outputs and guide prompt iteration
- Introduced session inactivity timeouts to prevent idle conversations and strengthen reliability
Graduate Student Researcher, CILVR, NYU
3D Computer Vision under Prof. David Fouhey · New York, NY, USA
January 2024 - May 2025
- Combined HaMeR transformer poses with COLMAP SfM to recover global 3D hand trajectories on Epic-Kitchens and EgoExo4D
- Optimized hand-motion trajectories using temporal smoothing with velocity and jerk regularization
- Recovered full-perspective 3D from weak-perspective predictions using SfM intrinsics and frame transforms
- Fine-tuned DINOv2 ViT-L/ViT-B and EfficientNet-V2 for hand-object contact classification on Hands23, ablating losses (Weighted/Balanced CE, Focal); best model reached 0.97 in-domain and 0.96 cross-dataset ROC AUC on Epic-Kitchens
- Evaluated Depth Anything and Metric3D as auxiliary depth signals for the hand-object contact task
- Developed visualization workflows with Rerun Viewer and scaled multi-GPU experiments on NYU HPC using SLURM
Course Assistant, NYU
Computer Vision, Deep Learning, LLVM courses · New York, NY, USA
September 2024 - May 2025
- Mentored 450+ students through office hours, grading, and project advising across 3 courses
- Assisted DL (Prof. Chinmay Hegde), LLVM (Prof. Saining Xie), and Advanced CV (Prof. David Fouhey)
B.Tech Thesis, PRIA Lab, IIT Indore
Multimodal Crowd Counting using Vision Transformers · Indore, MP, India
June 2022 - December 2022
- Trained a ViT-based model in a weakly supervised setting, achieving MAE 15.78 and MSE 30.70
- Designed a Pyramid Vision Transformer (PVT v2) model, achieving MAE 19.22 and MSE 32.94
- Ranked in the top 10 of 250+ B.Tech theses; nominated for Best Thesis award at IIT Indore
Research Assistant, NLP Lab, IIT Indore
Multi-Modal Hashtag Recommendation · Indore, MP, India
October 2021 - April 2022
- Modeled hashtag generation for micro-videos using LSTMs with self-, co-, and hierarchical attention
- Enhanced feature fusion with BERT, Word2Vec, ResNet152, InceptionV3, VGG16, MFCCs, and spectrograms
Projects
Kube SRE Gym
Awarded 1st place out of 100+ teams at the OpenEnv Hackathon (PyTorch, Cerebral Valley; $100K+ prize pool).
An RL environment where a small language model diagnoses and fixes live Kubernetes incidents on GKE via kubectl, with zero hardcoded knowledge of the cluster. An adversarial incident designer creates targeted scenarios based on tracked weaknesses, with curriculum control and an LLM judge scoring SRE workflow quality across expert personas.
Post-trained Qwen-series models with GRPO via TRL + vLLM on H100, using OpenEnv and Hugging Face Spaces. The agent learned to discover cluster topology, identify faults, and apply correct fixes from reward signal alone.
Point Tracking in Egocentric Videos
Course project under Prof. Mengye Ren for the Embodied Learning and Vision course at NYU. Evaluated self-supervised vision backbones (DINOv2, CroCoV2) zero-shot for egocentric point tracking on EgoPoints, then self-supervised fine-tuned both backbones with a lightweight delta network predicting per-frame keypoint displacement. Compared tracking accuracy across 8, 16, and 24 pixel thresholds.
Fine-Tuning Medical QA Models
Fine-tuned Meta LLaMA 3.2 3B for medical question answering using LoRA, QLoRA, and GaLore. Compared parameter-efficient methods head-to-head on USMLE Steps 1-3 and MedQuAD. Best USMLE accuracies: Step 1 42% (GaLore), Step 2 38% (QLoRA), Step 3 45% (GaLore); boosted MedQuAD F1 to 70% on open-domain medical QA — showing that efficient fine-tuning can meaningfully improve domain-specific performance on a 3B model.
VioletPass: Ticket Booking Platform
Architected a scalable ticket-booking platform on AWS that prevents double booking using Redis distributed locks. Configured Redis TTL-based seat reservations with PostgreSQL finalization for transactional integrity, plus QR-code verification for secure and efficient event check-ins.
AnimeVerse: Recommendation System
A personalized anime recommendation system that processes 300k user-anime interactions with PySpark. Compared collaborative filtering (ALS), content-based (TF-IDF), and semantic search (ChromaDB) approaches to surface relevant recommendations.
Education
New York University
Master of Science in Computer Science; GPA: 3.83/4
2023 - 2025
Course Work: Machine Learning, Operating Systems, Computer Vision, AI, Natural Language Processing with Representation Learning, Embodied Learning and Vision, Cloud Computing, Big Data.
Indian Institute of Technology (IIT), Indore
Bachelor of Technology in Electrical Engineering, Minor in Astronomy; GPA: 8.49/10
2019 - 2023