Open to chat and discuss new opportunities

Avneet Singh Sidhu.I build agentic AI systems.

AI software engineer at Katalyst AI, shipping production multi-agent systems for sales teams. Five startup stints across AI, full-stack, and on-chain — comfortable owning ambiguous product problems end-to-end with founders.

See my work Get in touch

Startup stints

AI · Web · Web3

Stack breadth

Where I've shipped

Experience

Five startup stints across AI, full-stack, and on-chain — sharpened the muscle of taking vague product ideas from scope to ship, working directly with founders.

AI Software Engineer
Current
Katalyst AI· NYC · Remote
Dec 2025 — Present
NYC-based AI startup building an agentic CRM copilot. Work directly with founders on product + engineering, owning features scope-to-ship.
- Built Home Chat, a multi-agent AI assistant orchestrating 25+ Zod-validated function-calling tools over Salesforce, Supabase, and Mem0 vector memory — with RAG retrieval, OpenAI and Gemini LLMs, and Vercel AI SDK SSE streaming for natural-language deal queries, follow-up actions, and designing the knowledge graph (Context Graph).
- Engineered Deal Intelligence, an OpenAI-powered agentic RAG pipeline auto-generating real-time CRM updates, email drafts, and signal-scored alerts across 10+ priority types from meeting + email events — owning prompt design, scoring heuristics, and product trade-offs with founders.
- Authored a fleet of agentic background services in Node.js — signals (Exa-powered hiring, leadership, and contact intent), research-agent (Perplexity-driven CRM field enrichment with citations), key-people (Google Vertex AI contact enrichment), and a daily rep-metrics cron — with bounded concurrency, idempotency, and Supabase pgmq queues.
- Owned the Gmail, Outlook, Fireflies, and Katalyst Notetaker integrations end-to-end via Composio MCP (OAuth, webhook triggers, sync); built gmail-sync — thread fetch with parallel triggers, atomic claim, and dual-pipeline EmailMemoryManager — cutting sync time by 3×.
- Shipped Pipeline Hygiene to production, a rule-based scoring engine (5 weighted rules, per-org penalties, time-series history) grading opportunity + rep health, and led a perf pass (Next.js Link prefetching, memoized SWR cascades, two-stage Salesforce contact enrichment) cutting page-to-page navigation by 50%.
Multi-agentRAGOpenAIGeminiFunction callingMCPVercel AI SDKSalesforceComposio MCPMem0Supabase (Postgres)Next.js 14TypeScriptPerplexityGoogle Vertex AIExa.jsp-limitZodpgmq
Software Engineer
BlockDAO· Remote
Jun 2025 — Nov 2025
Multi-chain crypto investment + onboarding tooling. Owned both the consumer-facing investment platform and the internal hiring system.
- Devised a multi-chain crypto investment platform with real-time NAV tracking, automated fee + price systems, and dual-tier user management — integrating Chainlink and CoinGecko price feeds with wallet-connect flows.
- Architected a job-onboarding system with inbuilt cognitive assessments, dynamic test evaluation, Slack + Calendly integration, and a unified admin dashboard powered by Supabase and React.
ChainlinkCoinGeckoSupabaseReactSlack APICalendlyWallet Connect
Software Engineer Intern
TravelGrowth AI· Remote
Mar 2025 — Jun 2025
Real-time streaming platform + WhatsApp bot. First production-scale React/Vue/Node experience with live peer-to-peer media.
- Engineered peer-to-peer live streaming via PeerJS WebRTC for Streamsell, a real-time streaming platform — shipping full-stack features across a React.js / Vue.js front-end and Node.js back-end with session management and REST API integrations.
- Shipped TravelQuest AI, a WhatsApp chatbot for travel queries — handling user dialog flow, message templates, and outbound notifications.
PeerJS WebRTCReact.jsVue.jsNode.jsWhatsApp Business API
Smart Contract Developer Intern
Taif Al Joud· Remote
Jul 2024 — Dec 2024
Solidity-heavy stint across three on-chain products. Picked up upgradeable patterns, gas optimization, and deep Hardhat tooling.
- Performed rigorous unit + integration testing of Solidity upgradeable contracts using Hardhat — focusing on OpenZeppelin libraries, ERC-20 / ERC-721, gas optimization, security, and inter-contract communication.
- Worked on smart contracts for a DEX, Launchpad, and NFT Marketplace — ensuring robust logic, debugging issues, and integrating with the frontend via Sepolia and Infura.
SolidityHardhatOpenZeppelinERC-20ERC-721SepoliaInfura
Blockchain Developer Intern
Blazpay· Remote
Jan 2024 — Feb 2024
Short, focused stint exploring account abstraction. First exposure to the operational rhythm of a founder-led team.
- Implemented account abstraction using Xion Blockchain and integrated Paymaster contracts to enable seamless cryptocurrency payments.
- Took part in daily strategic meetings with co-founders, contributing updates and presenting technical insights that fed into product decisions.
Xion BlockchainPaymasterAccount Abstraction

Things I built

Projects

Selected work I shipped end-to-end. The first is the deepest — six agents, a vector DB, and an MCP server all wired into one auditable pipeline.

Agentic Invoice Pipeline

Featured

Production-grade multi-agent invoice auditor with RAG, MCP, and a generative-UI chat

Live demo GitHub

A 6-agent LangGraph pipeline that ingests multilingual invoices (PDF, DOCX, OCR'd scans), validates them with deterministic rules, and produces auditable recommendations — pausing for human review when policy demands. Built end-to-end: backend, frontend, vector DB, MCP integration.

6-agent LangGraph with Postgres-checkpointed pause/resume for HITL approvals
Pydantic schema-validated structured outputs with auto-retry — no JSON parse hacks
ChromaDB Cloud RAG with 10 LLM function-calling tools, exposed via MCP to Claude Desktop
Live SSE pipeline telemetry visualised in a Next.js + ECharts analytics dashboard
Custom guardrails (PII redaction, prompt-injection block, off-topic gate) on chat I/O
LangSmith tracing across every agent invocation + LLM call

LangGraphLangSmithPydanticChromaDB CloudSupabaseMCPGroq Llama 3.3HuggingFaceFastAPINext.js 14Apache EChartsFramer MotionTesseract

Service Hub

MERN platform connecting users with on-demand service providers

Full-stack web platform with JWT-based authentication, role-aware APIs, scalable Mongoose data modeling, and secure media handling. Built to handle bi-directional matching between users and providers across categories.

JWT-based authentication with role-aware route protection
Mongoose schemas optimized for category filtering and provider discovery
Image upload and validation pipeline with size + type gates
React front-end with Axios-driven API calls and responsive layouts

ReactNode.jsExpressMongoDBMongooseJWT

What I work with

Technical Skills

Grouped by where they live in my day-to-day. Everything here is something I've shipped with — not just read about.

AI / LLM

LangGraphLangChainLangSmithMCPRAGAgentic RAGVercel AI SDKMem0Pydantic AIFunction callingStructured outputsPrompt engineeringHITL workflowssemantic-chunking

Models & Embeddings

OpenAI APIGroq (Llama 3.3)Google Vertex AIPerplexity APIHuggingFacesentence-transformers

Vector DB & Storage

ChromaDB CloudSupabase (Postgres)pgmq queuesMongoDBSQLAWS S3IPFS

Languages

TypeScriptJavaScriptPythonSolidityRustC/C++

Frameworks

Next.js 14ReactNode.jsExpressFastAPIZodHardhatWeb3.jsEthers.jsAnchor

Infra & Integrations

Composio SDKWebhooksOAuthSSE Streamingnode-cronExa.jsSalesforceSlack APICalendlyWhatsApp Business APIGitGitHub Actions

Web3

EthereumSolanaSmart contractsDeFiDAppsNFTsAccount Abstraction

Soft skills

Founder collaborationScope → ship ownershipDaily code reviewFast-paced startup paceClear written + verbal communication

Reach out

Let's build something.

Open to founding-engineer and deep-AI roles. Comfortable working remote, happy to relocate, fast at replying.

avneetsidhu18@gmail.com

avneet-singh-sidhu

GitHub

Sidhuavneet

Designed and built by Avneet · Next.js 14 · Tailwind · Framer Motion