
Hello, I'm
Kaustubh
Patange
Building high-performance backend systems, AI-powered platforms, and contributing core numerical computing to open source.
/01Where I've Worked
Experience
Google Summer of Code Contributor
@stdlib-js/stdlib- ◆Extending Level 2 and Level 3 BLAS routines (real and complex) in JavaScript, C, and Fortran for near-native performance in Node.js.
- ◆Implementing a three-layer API architecture (CBLAS-compatible interface, strided-array compute kernels, and ndarray wrappers) with loop unrolling and cache-friendly memory-access optimizations.
Core Contributor
@stdlib-js/stdlib- ◆Shipped 200+ merged PRs spanning BLAS routines, statistical reductions, and native C add-ons, each with full test suites, benchmarks, and documentation.
- ◆Designed 60+ ndarray API packages and numerically stable computation kernels in JavaScript and C; reviewed community PRs to enforce strict API design, testing, and documentation standards.
Software Developer
@Nexentia- ◆Built backend services and secure REST APIs for a Warehouse Management System, with indexed relational schemas, ACID transactions for stock consistency under concurrent orders, and JWT role-based access.
/02What I've Built
Featured Projects
LLM Inference Engine
Built from Scratch on Apple Silicon
Engine serving Qwen3.5-2B (hybrid Gated DeltaNet + attention) in INT4 on an 8 GB MacBook Air via an OpenAI-compatible API with continuous batching, paged KV cache, and chunked prefill. Hand-written Metal kernels (quantized matvec, paged attention, batched DeltaNet step) raised 8-stream decode from 43 to 116 tok/s, 1.6x MLX-LM and 1.9x llama.cpp on the same machine. Verified against Hugging Face Transformers to ≤3.8e-6 per layer in fp32, with calibrated mixed-precision INT4 at KL 0.045 and 700+ automated checks.
32 Bytes
Embedding Search with Cohere Embed 5
Studied retrieval at 8 KB (float), 1 KB (int8), and 32 B (binary) per page with a two-stage search (binary first pass, int8 re-check), evaluated with RCP-nDCG on ViDoRe v3 using clustered bootstrap CIs and Holm correction. Hand-wrote Metal kernels for exact top-100 over 10M binary codes in 4.45 ms (1.2x FAISS, up to 4.3x batched) at 78 GB/s, and found and reported 3 reproducible issues in Cohere’s open-source rcp-ndcg evaluation library.
Corpora
Multimodal Agentic RAG Platform
Multi-tenant agentic RAG pipeline using LangGraph with hybrid vector search (dense + BM25, RRF fusion) and cross-encoder reranking — streaming citation-grounded answers over SSE at 31ms p95 retrieval latency. Async ingestion, CLIP text-to-image search, CI-gated retrieval evals, Clerk JWT multi-tenancy, and Langfuse observability.
CoSketch
Real-time Collaborative Whiteboard
Low-latency real-time collaboration system using persistent WebSocket connections to synchronize canvas operations across multiple concurrent users. Event-driven canvas state architecture for drawing, shape transformations, and persistent workspace management.
/03Client Work
Freelance Work
Aakar Design Studio
Freelance Engagement
Architected a CMS-backed studio portfolio end-to-end using Next.js 15 and Sanity, empowering non-technical staff to seamlessly manage the homepage, project catalogs, and media galleries. Owned SEO and serverless deployment via Cloudflare Workers (OpenNext) with KV edge caching — dynamic sitemaps and JSON-LD structured data for full search indexability.
/04My Toolbox
Technical Skills
Languages
Frameworks
AI & ML
Tools & Infra
/05Learning & Recognition
Education & Achievements
Education
Walchand Institute of Technology
Solapur, Maharashtra, India
B.Tech in Computer Science and Engineering
Minor in Electronics & Telecommunication Engineering
Related Coursework
◆ Linear Algebra
◆ Digital System Design
◆ Database Management Systems
◆ Scientific Computing
◆ Distributed Systems
◆ Data Structures & Algorithms
LeetCode
ProfileSolved 200+ data structure and algorithm problems.
stdlib Member
RepositoryMember of the numerical library stdlib on GitHub.
Anthropic
AnthropicPart of Anthropic's Claude Open Source Program.
Cohere Labs
Cohere LabsActive member of the Open Science Community.
/06Let's Connect
Get In Touch
Open to internships, freelance work, open source collaboration, and interesting engineering problems. Inbox always open — I'll get back fast.