Open source
cachellm
Wraps your Anthropic, OpenAI or Gemini client and sets up prompt caching for it automatically. On npm and PyPI.
- Role
- Type
- Status
- Licence
- Stack
Hi, I'm Sahil
From LLM orchestration to model security: the systems that keep AI products fast, cheap and safe.
Capabilities
LLM Systems
Orchestration, tool calling, memory, and RAG pipelines that turn a model into a product, with cost and latency budgets built in from day one.
Agent Workflows
Multi-agent architectures for enterprise use cases, owned end to end from research and prototyping to production deployment and monitoring.
AI Security
Red-team testing, prompt-injection evaluation, and security-specialized models built for authorized penetration testing and defensive research.
Full-stack Apps
Next.js, TypeScript, and Python applications with Supabase, Stripe billing, and Dockerized backends on AWS or GCP.
Data Engineering
ETL pipelines, dataset curation, and evaluation harnesses that keep training data clean and models honest.
Selected work
- Research
RedSec
Security-specialized LLMs fine-tuned as LoRA and QLoRA adapters for authorized red-team work. Models and datasets on Hugging Face.
- Role
- Type
- Published
- Hardware
- Stack
- Writing
MCP Apps Sandbox
An evidence-based walkthrough of where an MCP App's HTML runs inside Claude.ai and what it can reach.
- Role
- Type
- Finding
- Includes
- Stack
Open source
Fixes I have sent upstream to projects I use, and the repositories I maintain in the open.
Upstream contributions
eclipse-theia/
theia Cloud and desktop IDE framework
Fixed remote edits landing at the wrong position in live collaboration
MergedOct 2026
When a collaborator's change contained more than one edit, every edit after the first insert or delete was applied at a displaced offset. I corrected how the Yjs delta is walked against the document before the edit, moved it into its own module and added tests.
View pull requestanthropics/
claude-code-action Claude Code for GitHub Actions
Made the label trigger match without regard to case
MergedAug 2026
A workflow set to trigger on one label did not fire when the label on the issue differed only in capitalisation. GitHub treats label names as unique without regard to case, so the comparison now does too, with a test covering it.
View pull requesteclipse-theia/
theia Cloud and desktop IDE framework
Stopped the metrics endpoint emitting invalid output at startup
In reviewOct 2026
For the first ten seconds after the backend starts, the metrics endpoint included the literal text undefined, which is not valid Prometheus format. The fix initialises the plugin metrics and guards the concatenation, with tests for both.
View pull requestFreedomScientific/
standards-support Screen reader support tests for web standards
Added aria-level coverage to the HTML test page
MergedMay 2023
A small addition to the suite that tracks how JAWS handles web standards.
View pull requestmdn/
interactive-examples Live code examples on MDN Web Docs
Completed the tab-size example for the CSS reference
MergedMar 2021
My first merged pull request to a widely used project, reviewed by the MDN team.
View pull request
Also maintained
- nifty-options-research-labA backtesting lab for Indian index options that publishes its negative results: the strategies that lose, and the tooling that showed why.View repository
- groww-market-dataDaily price, volume and open interest data for Indian markets, with the collector that gathers it.View repository
Experience
Neural Arc Inc
Pune, on-site
Lead of Cognitive Intelligence
Jul 2025 - Present
Designing generative AI systems and multi-agent architectures for enterprise customers and owning them from prototype to production: LLM orchestration, tool calling, memory, RAG, Dockerized backends on AWS and GCP, and usage-based billing with Stripe.
Quantum Coder
Mar 2025 - Jul 2025
Core LLM features and agent workflows for early-stage AI products, with the backend APIs and cloud deployments behind them.
Ampersand
Pune, on-site
SDE I
Mar 2025 - Present
AI-powered web and app features across the stack with React, AWS and modern backend frameworks. Promoted from intern on the strength of delivery and ownership.
AI and App Development Intern
Dec 2024 - Feb 2025
Frontend components, backend APIs and cloud services for AI-integrated application features.
LvlUp Ventures
Remote
Venture Scout
Oct 2025 - Present
Sourcing and evaluating early-stage startups for the fund.
IndiaResults.com
Jaipur, on-site
Data Engineer Intern
Jun 2023 - Aug 2023
Data processing and ETL pipelines on production datasets.
Fastlegal
Jaipur, remote
Data Engineer Trainee
Jan 2023 - Feb 2023
Data integration and ETL work.
Jaypee University of Engineering and Technology
B.Tech, Computer Science, 2020 - 2024
- Oracle Agentic AI Certified Foundations Associate
- Goldman Sachs Software Engineering Virtual Program, 2023
Profile
With more than two years of shipping AI products, I focus on LLM orchestration, agent workflows, and the infrastructure that keeps them fast, cheap, and safe. I lead Cognitive Intelligence at Neural Arc and build open-source tools like cachellm and RedSec. Let's build something incredible together!
Works with
- Large language models
- LLM orchestration and agent workflows
- Retrieval-augmented generation
- Prompt caching
- AI security and red teaming
- Fine-tuning with QLoRA
- Next.js and TypeScript
- Python
- Data engineering
Contact
Open to collaborating on AI infrastructure, agents, and security tooling.
