shreshth kharbanda.

$ cat

skills.txt

Core Languages: JavaPythonC++
ML & Inference: PyTorchTensorFlowvLLMFlashAttentionSageMakerTritonTransformers
Cloud & Infrastructure: AWS (ECS, Fargate, API Gateway, Lambda, SageMaker, Bedrock, CDK)DockerTerraform

$ git log

6 roles

June 2023 — Present

Launched a new API introducing guardrails for agents with a tool adherence policy. Interacted with 50+ customers to scope and drive integrations enabling safe production usage of agentic AI.
Launched Bedrock Guardrails Standard Tier in partnership with research and product driving 30%+ accuracy gains, 25% higher GPU utilization via cross-region inference, expansion to 60+ languages.
Defined and owned a model-agnostic integration framework reducing model onboarding from 17 days → 1 day and enabling 40+ model launches at re:Invent.
Re-architected stream buffering and evaluation improving time-to-first-token by 67% with equivalent accuracy.
Architected tier-aware admission control to mitigate noisy-neighbor effects during burst traffic, improving service availability by 15% under load shedding leveraging Redis + Lua.
Designed and implemented multi-endpoint slot-pooling architecture and dynamic batching for inference improving throughput by 3x and reducing p95 latency by 50%.

January 2023 — June 2023

Implemented features for Tesla’s mobile app serving 60k+ weekly users, partnering with product and design.
Led development and launch of a customer-facing installer discovery service supporting 20k+ weekly users.
Implemented Redis-based caching for the backend-for-frontend, reducing page load latency by 65%.

April 2022 — June 2023

Internationalized a research-backed educational social media platform with 500,000+ users enabling scalable, privacy-compliant data collection.
Built distributed data pipelines and supervised learning models (regression + classification) to identify engagement bottlenecks and predict high-friction modules, reducing manual analysis by 1,000+ hours.

June 2022 — August 2022

Applied incremental static regeneration for a micro-frontend to reduce load speeds by 300% for 20M+ users.

June 2020 — August 2020

Developed an Android misinformation-detection app using Java and REST APIs, tested with 30+ beta users.
Designed and trained a TensorFlow NLP model for bias detection achieving 77% accuracy on labeled news article data.

April 2020 — June 2022

Drove 25+ digital transformation projects using React.js, React Native, Java, Node.js, Python, AWS, GCP.

$ ls

3 side bets

Built a personal agent that acts as an executive assistant interfacing through iMessage using Java SpringBoot + MCP.

Built a long-running AI agent that operates continuously across development and operational workflows, autonomously planning work, implementing changes, reviewing code, and investigating using persistent memory and structured code context.
Automates 25+ hours/week and reduces on-call triage from 30 → 7 minutes with 85% root cause accuracy.
Produces PRs that are 40% mergeable on first revision and converge within ~3 iterations on average.

Built a search engine in C, C++ with dynamic indexing processing 10,000+ documents.
Optimized tool using multithreading, big-endian handling, and architecture-neutral data marshalling.
Enabled secure and concurrent request handling using Boost, STL to reduce server response time by 30%.

$ cat

Coursework: Distributed Systems, Hardware Systems, Deep Learning, AI, ML, Data Structures & Algorithms, Software Design, Databases.
Organizations/Clubs: Algorithmic Trading Club, DubHacks, Computing Community.