Agrim Jaimini

• Built a generalized GitHub API proxy in Go as shared developer infrastructure, unifying high-concurrency GitHub access behind a single internal API for autonomous CI systems and company-wide tooling
• Engineered a Redis cache keyed on ETags and conditional requests to absorb high-concurrency agent traffic, keeping agents under GitHub's primary rate limits where naive per-agent access would exhaust them
• Added Datadog metrics, dashboards, and error-rate alerting to track proxy health and per-agent traffic

May 2026 – Aug 2026 · San Francisco, CA

• Delivered end-to-end solution for BountyX funding platform built on XRP Ledger through XRPL Builder Residency, enabling secure on-chain payouts for open source contributions and 100+ bounty transactions
• Architected REST microservices with FastAPI, PostgreSQL, and xrpl-py for bounty creation, claim validation, and automated settlement; integrated GitHub API to verify merged PRs and prevent fraudulent claims
• Deployed on AWS EC2 with Docker, implementing CI/CD pipelines, comprehensive unit tests, and structured logging to ensure payout traceability and audit compliance for enterprise-grade financial operations

Jul 2025 – Aug 2025 · New York City, NY

• Designed and deployed modular ML pipeline processing 1M+ data points with event-driven architecture, AWS Lambda ingestion, FastAPI inference service, and Next.js monitoring dashboard; achieved 99% uptime through containerized deployment with Docker on AWS EC2 and CI/CD automation
• Built scalable model versioning system with PostgreSQL enabling A/B testing across 10+ iterations, reducing deployment rollback time by 75% and improving experiment reproducibility for production ML systems
• Trained PointNet deep learning model using PyTorch achieving 93% accuracy with <300ms inference latency; implemented synthetic data augmentation increasing training data by 5X to improve model robustness

May 2025 – Jul 2025 · Dallas, TX

• Built Ethereum transaction dashboard with NetworkX and PyVis processing 1M+ daily transactions for real-time wallet-network exploration, fund-flow visualization, and volume-based filtering to support fraud investigation workflows
• Integrated anomaly detection with Isolation Forest and Louvain community detection to identify suspicious wallets and cluster related addresses, improving fraud analysis precision and reducing manual investigation time

Oct 2024 – Jan 2025 · Remote

• Built a trace-driven Rust simulator modeling transformer inference (prefill/decode) on configurable hardware, estimating latency from compute and bandwidth costs to classify workloads as compute- or memory-bound
• Added parameter sweeps, plots, and CI regression tests to compare hardware configurations (compute throughput, memory bandwidth, interconnect) and quantify their impact on end-to-end inference latency

Rust · Python

• Built a runtime that splits software specs into dependency-aware parallel tasks, runs workers in git worktrees, and gates merges on automated and human review, with Claude Code and Codex as pluggable backends
• Designed graph + vector memory so agents learn across tasks: past runs, decisions, and review outcomes are embedded and entity-linked in SQLite, then recalled via similarity search and multi-hop graph expansion into each agent's prompt; exposed agent tooling over an MCP server consumed directly by Claude Code

Python · TypeScript · Next.js · SQLite · MCP

• Built a shared research-memory layer where AI agents cache and resell deep-research results, ranking memories by semantic similarity, freshness decay, and LLM-judge quality to serve a cached hit or trigger re-research, with Arize tracing and evals; agents pay per retrieval via x402 machine-to-machine micropayments

TypeScript · Next.js · Vercel AI SDK · pgvector · Arize

• Engineered preprocessing for prompt/chosen/rejected triples with preference dataloaders and DPO loss computation
• Custom SFT + DPO PyTorch trainers with bfloat16, gradient checkpointing, and tokenizer/model consistency checks
• Colab-friendly configs and evaluation suite covering preference accuracy, reward margin, perplexity, and sample generations

PyTorch · Hugging Face Transformers · DPO · Google ColabGitHub ↗

• Ranked tickets by severity using embeddings, duplicate signals, and component history with PyTorch + LightGBM on GKE
• Event-driven ingestion with Kafka plus Redis caching for sub-200ms reads and explainable REST endpoints
• Next.js dashboard surfacing severity, duplicates, and model explanations for engineering teams

Next.js · Node.js · Kafka · PostgreSQL · Redis · PyTorch · Kubernetes · GCPGitHub ↗

• Semantic search across 1k+ documents with sentence-transformers embeddings and k-means clustering (95% relevance)
• React + Express full stack with embedding-powered retrieval and topic grouping
• Iterated clustering via silhouette score tuning to accelerate knowledge discovery

Python · React · Node.js · MongoDB · sentence-transformersGitHub ↗

• Data Structures and Object-Oriented Programming (Java)
• Functional Programming and Advanced Data Structure (OCaml)
• Analysis of Algorithms
• Machine Learning
• Database Systems
• Discrete Math
• Linear Algebra
• Backend Development
• Blockchain Technology
• Computer Architecture
• Probability, Vectors, and Matrices in Computing

4.04 GPA · Expected May 2028