Headline

Applied AI Engineer | LLM gateways, agent orchestration, inference | Author of Maya OS (npm) | Research Intern @ MISN Lab, IIT Delhi

About

I build the layer underneath the models: LLM gateways, agent runtimes, and inference that runs on your own machine. At Colate I built the multi-provider LLM gateway behind Cana. One shared request/event representation translates OpenAI, Anthropic and Gemini formats, covering streaming, tool calls and multimodal inputs. Prompt caching and token-usage accounting took it to 6.05B cached input tokens, 53% input-token reuse and 35% lower inference latency. I also built its multi-agent orchestration, with worker, reviewer and lead roles, dependency-aware delegation, inter-agent messaging and completion handoffs for long-running tasks, plus persistent AWS and Kubernetes runtimes with session recovery. I am also the author of Maya OS, a local-first AI operating system published on npm. One command boots an OpenAI and Anthropic compatible gateway that routes across cloud providers and local GGUF models via llama.cpp and Metal, with RAM guardrails that degrade instead of crashing the host. On top of it: agent teams running on real PTYs, a message bus, an orchestrator and a dashboard, across an 8-package TypeScript monorepo. npm i -g maya-os Alongside production work I research agentic systems at the MISN Lab, IIT Delhi, under Dr. Sandeep Kumar, and train small models from scratch to understand how they fail: F75, a 100k-parameter LM trained on a single RTX 5060 Ti, and LlamaVision, a 6.3M-parameter vision projector for a frozen Llama-3.3-1B. What I work with: model gateways, MCP, multi-agent orchestration, RAG and vector databases, inference optimization, llama.cpp and GGUF, PyTorch, FastAPI, Docker, Kubernetes, AWS, Python and TypeScript. Hackathon wins: HIVE at IIT Delhi 2024, FinAThon at IIT Delhi 2023, top 10 internationally at IIT Bombay TechFest 2023. CV: https://himansh.in/r Site: https://himansh.in Maya OS: https://maya.himansh.in

Location

Bengaluru

AI stack

claude code

codex

ollama

huggingface

llama.cpp

PydanticAI

Work experience

Applied AI Engineer at Colate, Jul 2025 - Present · 1 yr 4 mos, United Kingdom, Built the multi-provider LLM gateway and agent runtimes behind Cana. • Developed a multi-provider LLM gateway on a shared request/event representation that translates between OpenAI, Anthropic and Gemini formats, covering streaming, tool calls and multimodal inputs. • Implemented prompt caching and token-usage accounting: the internal dashboard reported 6.05B cached input tokens (53% input-token reuse) and 35% lower inference latency from caching and runtime optimizations. • Built multi-agent orchestration for Cana with worker, reviewer and lead roles, dependency-aware delegation, inter-agent messaging and completion handoffs for long-running tasks. • Engineered a vector database and semantic code-search engine, indexing repository-scale corpora for sub-second retrieval. • Shipped persistent AWS and Kubernetes runtimes with session recovery so long-running agent jobs survive restarts without losing context. • Built an evaluation harness across 90+ test prompts to catch robustness regressions before release.

Research Intern at Indian Institute of Technology, Delhi, May 2026 - Aug 2026 · 4 mos, Hybrid, • Engaged in research activities on “Agentic Systems for Recommendations” at IIT Delhi. • Collaborated with faculty and peers to contribute to ongoing projects in the Department of Electrical Engineering. • Adhered to academic and administrative guidelines while receiving a consolidated stipend of ₹15,000 per month.

Technical Lead at Vorcus AI, May 2025 - Jul 2025 · 3 mos, India

SDE Intern at Indian Institute of Technology, Delhi, Oct 2024 - Oct 2024 · 1 mo, On-site, I was selected for a Software Development Internship at the prestigious Vision Lab, IIT Delhi. I began working on the assigned tasks and contributed actively for over a week. However, due to NOC (No Objection Certificate) constraints from my university, I was unable to continue the internship officially. Despite the short duration, it was a valuable learning experience that gave me exposure to high-level research and development workflows in computer vision and AI.

Django Developer Intern at ID8NXT, Jul 2024 - Sep 2024 · 3 mos, Remote, Working with the senior development team on backend features, APIs, and data models. Enhancing functionality, debugging issues, and applying best practices in Python and Django. Gaining hands-on experience with remote collaboration, version control (Git), and modern software development workflows.

Education

Bachelor of Technology - BTech, Computer Science, IILM University, Greater Noida

Public links

github: https://github.com/iamthehimansh

codeforces: https://codeforces.com/profile/iamthehimansh

himansh.in: https://himansh.in/