I build backends end-to-end — distributed systems, LLM gateways, middleware automation, and the infrastructure that keeps them reliable. Backend developer focused on microservices, cloud architecture, and open-source work in the AI tooling ecosystem.

Maintainer @ InfiniFlow — dedicated maintainer of the MinerU PDF/layout parser in RAGFlow (90K+ GitHub stars). Own MinerU-related issues and PRs; contributed MonkeyOCR as a dedicated PDF backend (#19044).

Products & systems

• OPENINFER — self-hosted LLM gateway with routing, guardrails, retrieval, budgets, tracing, evaluations, and an admin console. Includes oi, a hardware-aware CLI for finding and running local open-source models. Live at openinference.tech.
• netsuite-middleware — AWS middleware automation. Microservices with gRPC, event-driven workflows, PostgreSQL, Redis, and RabbitMQ for NetSuite integration at scale.
• netssuite — production-ready AI knowledge-base assistant for NetSuite. Multi-client deployment with dynamic URL generation and an embedded chatbot layer.
• NETFLIX_GPT — AI-driven Netflix discovery with natural-language search and personalized recommendations.

Backend & APIs

• API_Gateway_Flights — flight booking API gateway with service routing and aggregation.
• axio-backend — Node.js backend services for product APIs.
• dsatrackerllmbackend — LLM-backed DSA progress tracking backend.
• telejob — job-hunting bot that filters LinkedIn and Telegram posts with regex + Gemini, then tracks applications in Telegram.

Learning & tooling

• LEETCODE-DSA — curated LeetCode solutions in C++.
• AlgorithmPractising — algorithm practice workspace in TypeScript.
• DEV-NOTES — Obsidian-synced notes on Go, distributed systems, and backend patterns.
• llm-bot — JavaScript LLM bot experiments.

50 merged upstream PRs plus active work across LLM platforms, local inference, agents, and ML infrastructure. All pull requests.

LLM platforms

• infiniflow/ragflow — maintainer, MinerU parser (90K+ ★). 2 merged: dataset navigation tree load failures (#17668), empty canvas autosave guard (#18783). Open: MonkeyOCR PDF backend (#19044), reverse-proxy subpath deployment (#18890), wiki retrieval fields (#18787), tenant model-type migration (#18781).
• langgenius/dify — 6 merged: legacy model_type cleanup (#39708), slash variable picker guard (#39761), cached recommended-app template fetches (#40176), UTF-8 binding-file preview (#41761), boolean server defaults (#41762), unclosed think tags around agent tool calls (#41764). Open: WebApp file input defaults (#41763), Swagger API tool credential persistence (#41625), configurable indexing workers (#41302).
• Mintplex-Labs/anything-llm — merged: You.com web search for agents (#6058). Open: generic OpenAI sampling passthrough (#6048), saved endpoint connections (#6065).
• BerriAI/litellm — 2 merged: form-encoded video edit/extension bodies (#36513), managed batch output files (#34092). Open: internal metadata leak (#36412), Prometheus unmatched-path labels (#36410), system_prompt logging (#36406).
• langfuse/langfuse — open: monitor alert filter evaluation (#16719), comment-mention email persistence (#16580), login rate limits (#16577), observation IO in legacy trace downloads (#15451), admin API to list orgs by email (#15453).

Agents & orchestration

• mastra-ai/mastra — 2 merged: sequential allow-tool results after auto-approved resume (#19940), Zod v4 jsonSchema adapter (#21187). Open: publish stored skill drafts (#23032), deployment thinking defaults (#23031), workspace resolution from project root (#22861).

Local training & inference

• unslothai/unsloth — 34 merged across Studio, installer, and inference. Recent merges include llama.cpp update rate-limit handling (#9980), empty Downloads list entry (#9849), Xet stall HTTP resume (#10045), custom TTS endpoint repair (#9905), and local agent provider routing (#9871). Open: API tokens on plain HTTP (#9906), empty web_search recovery (#9716), FP8/FP4 compressor consent (#9554). All Unsloth PRs.
• unslothai/unsloth-zoo — open: forward enable_thinking=False in the Qwen3.8 vision collator (#1072).
• langgenius/dify-official-plugins — 3 merged: parse plugin files when Dify already has an event loop (#3756), Cohere embed-v4.0 models (#3769), Tongyi temperature floor (#3767). Open: qwen3.8-flash model support (#3822), workspace-specific DashScope API base (#3789), Anthropic workspace identity header (#3788).

ML pipelines

• kubeflow/pipelines — open: in-cluster KFP client DNS (#13820), parameterized container images (#13821), cyclic dependency detection (#13823), pipeline I/O validation (#13824), default StorageClass for CreatePVC (#13834).

Node.js · TypeScript · Python · Go · Java · Spring Boot · gRPC · PostgreSQL · Redis · RabbitMQ · MongoDB · AWS · Docker · Kubernetes · React · Next.js · Fastify · Dify · Langfuse · LiteLLM · RAGFlow · MinerU · Unsloth

• Portfolio — souravportfoliov1.netlify.app
• LinkedIn — linkedin.com/in/souravrajvi
• Email — souravrajvi@gmail.com
• GitHub — github.com/Souravrajvi0