Deep Trivedi: Software & AI Engineer

Open to freelance & full-time roles

I build

I build fast, reliable systems: LSM-tree storage engines in Rust, multi-tenant MCP servers, production RAG pipelines.

Hire me See the work

$ ./deep --bootv2026.07

[0.000s] init deep_trivedi · software & ai engineer

[0.017s] mount cairn (lsm-tree kv) · reads 17× faster

[0.042s] serve mcp · stateless · multi-tenant

[0.108s] tune rag · +25% retrieval accuracy

[0.230s] ready · open to freelance & full-time

Scroll↓

RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·

AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·AI AGENTS·VECTOR DB·RAG PIPELINES·LANGGRAPH·RUST·DISTRIBUTED SYSTEMS·MCP SERVERS·RAG PIPELINES·STORAGE ENGINES·

01/ WORK

Where my code met production.

Internships and open source with real users on the other end - AI infrastructure, consensus protocols, and the CI that keeps them honest.

Dec 2025 - Present

Bengaluru, KA

Current

Humantic AI

Software Engineering Intern

—Designed and shipped a stateless multi-tenant MCP server exposing the product as LLM-callable tools, with a token-based auth layer integrating multiple internal APIs - backward compatible by construction.
—Architected a serverless AWS Lambda monitoring service tracking third-party API quotas within 80% and triggering real-time Slack alerts, preventing service blackouts and ensuring 24/7 reliability.
—Optimized a RAG pipeline with semantic chunking and metadata filtering - a 25% improvement in retrieval accuracy and fewer LLM hallucinations.

LangGraphRAGPythonNode.jsReact.jsMySQLMongoDB

Oct 2025 - Nov 2025

Remote

Keploy.io

Software Engineering Intern

—Developed and optimized GitHub Actions pipelines to resolve workflow conflicts and improve CI/CD.
—Hardened the PostgreSQL parser with safety pre-checks to prevent runtime panics, and tuned CLI/GUI features to cut system latency.

GolangDockerNode.jsGitHub Workflows

Jun 2025 - Sep 2025

Remote

Linux Foundation (LFX '25)

Mentee, Hyperledger Fabric

—Integrated the BDLS consensus protocol into the Orderer node, strengthening Byzantine Fault Tolerance and tuning goroutines to raise transaction throughput and cut latency.
—Collaborated with core maintainers on architectural design and code reviews inside a large-scale blockchain system.

GolangBFT ProtocolDistributed Systems

Aug 2024 - Apr 2025

Remote

Open Source

Contributor

—stdlib-js: implemented static functions, fixed CI/CD workflow issues, and resolved multiple bugs to improve the JS standard library's stability.
—Tor Network: built a bridge-pool indexer to parse data from Java into Rust (under a GSoC '25 issue).

JavaScriptRust

02/ FLAGSHIP / BUILT AT HUMANTIC AI

One server, every tenant, zero shared state.

A stateless, multi-tenant MCP server

I designed and shipped a Model Context Protocol (MCP) server that exposes Humantic's product as a clean set of LLM-callable tools, so any AI agent can use it like a native capability.

01Stateless & multi-tenant by design: one server safely serves many tenants with no shared state.
02Token-based auth layer that fans out across multiple internal APIs.
03Backward-compatible by construction, so existing integrations never break.

MCPLLM tool-callingNode.jsPythonAuthMulti-tenancy

tenant_a ──┐
tenant_b ──┼──▶  [ mcp_server ]  ──▶  internal APIs
tenant_c ──┘        stateless
                    token-auth
                    zero shared state

03/ PROJECTS

Built from scratch, then benchmarked.

I build the engine instead of importing it - a storage engine, a search engine, a realtime app, an on-chain program. Numbers included.

/01

Cairn

Persistent key-value store in Rust on an LSM-tree.

A from-scratch storage engine: write-ahead log, sorted memtable, and immutable on-disk SSTables. An in-memory sparse index plus per-table Bloom filters cut read latency 17× by skipping redundant disk seeks for absent keys. Parallel background compaction merges SSTables and reclaims deleted space.

faster reads

WAL

crash-safe

RustLSM-treeBloom filtersWALLinux

View source

/02

Mini Search Engine

Multithreaded search engine in C++ with an inverted index.

Inverted index plus a trie for fast keyword lookups. Tokenization and indexing are parallelized to clear throughput bottlenecks, and relevance is sharpened with stop-word removal, normalization, and BM25-inspired scoring.

BM25

ranking

Multi-core

indexing

C++MultithreadingInverted IndexBM25Trie

View source

/03

Talkie

Real-time chat app on the MERN stack.

A full-stack chat application with live messaging over Socket.io, MongoDB persistence, and a React front end.

Realtime

sockets

MERN

full-stack

ReactNode.jsExpressMongoDBSocket.io

View source

/04

Blog DApp

Decentralized blog on Solana with Anchor.

A blogging dApp on Solana, built with the Anchor framework: on-chain programs paired with a JavaScript client.

Solana

on-chain

Anchor

programs

SolanaAnchorRustWeb3JavaScript

View source

Five defaults I bring to every system.

/01

I go one layer below the abstraction.

APIs lie; source code doesn't. When something breaks, I read the implementation, not just the docs.

/02

Correct first. Then I make it fast.

A 17× speedup means nothing if the answer's wrong. I earn performance only after I earn correctness.

/03

Benchmarks over vibes.

I don't guess where the time goes. I measure, profile, and let the numbers pick the fight worth having.

/04

I'd rather build the engine than import one.

Storage engines, search indexes, consensus: understanding the hard part beats gluing libraries together.

/05

Boring in production is a feature.

Stateless, backward-compatible, observable. The exciting part should be the product, not the incident channel.

05/ STACK

Tools, school, and the rating grind.

Languages

RustC++GoPythonTypeScriptJavaScriptJavaCSQLBash

AI / ML

LangGraphLangChainRAGAgentic WorkflowsMCPLangSmithChromaDB

Backend & Web

Node.jsExpressNext.jsReactDjangoSocket.iogRPC

Data & Infra

PostgreSQLMongoDBMySQLRedisAWSDockerCI/CDLinux

Education

Indian Institute of Information Technology (IIIT)

B.Tech, Computer Science & Engineering

Nov 2022 - Jun 2026 · Jabalpur, Madhya Pradesh

Coursework and obsession across systems, distributed systems, DBMS, networks, operating systems, and DSA, paired with a heavy competitive-programming habit.

Competitive programming

1000+ problems solved

06/ CONTACT

Let's build the part your product quietly depends on.

Storage, retrieval, agent plumbing: the layer nobody notices until it breaks. That's the layer I'm best at.

GitHub LinkedIn LeetCode Codeforces

Open to freelance & full-time roles · Usually replies within a day