LinkedIn
X
back to portfolio

what i build

Everything I ship as a forward deployed engineer. Fixed-scope sprints or embedded retainers — remote, worldwide.

book a free callemail me
01

Forward Deployed AI Engineering

I join your team for a scoped engagement, learn the workflow from the people who run it, and take one AI use case from messy reality to production — inside your stack, under your security rules.

shipped: Machine Maintenance Bot (live on a factory floor) · SiteOS (live for an architecture studio)

  • Discovery with the actual end users, not just the sponsor
  • Working pilot in your environment within the first weeks
  • Production rollout: auth, monitoring, fallbacks
  • Handover docs and a team that can run it without me
02

AI Agents, Chatbots & WhatsApp Bots

Assistants that answer from your own content instead of hallucinating — on your site, in Slack, or on WhatsApp where your customers already are. Multi-step tool use and a clean handoff to a human.

shipped: Machine Maintenance Bot · SiteOS · Chatter AI · LawGPT

  • Embeddable support & sales chatbots
  • WhatsApp Business API agents
  • Multi-step agents with tool calling
  • Guardrails, evals and fallback handling
03

AI Voice Agents

Phone agents that hold a real conversation in Hindi and English — qualifying leads, booking appointments and handling inbound calls without a queue.

shipped: Mona @ RnDynamos Labs · bolna-ai/bolna · ArduPilot Assistant

  • Inbound & outbound calling agents
  • Telephony integration (Twilio / Plivo)
  • Call transcripts, summaries and CRM sync
  • Latency and interruption tuning
04

Enterprise Integrations

Most of forward-deployed work is plumbing. I connect AI to your CRMs, drives, databases and SaaS tools with auth, retries and error handling that survive production.

shipped: SharePoint ticketing in Machine Maintenance Bot · SiteOS SharePoint archive · archestra-ai connectors

  • SaaS & internal API integrations (OAuth 2.0)
  • Notion, SharePoint, OneDrive & Drive connectors
  • Slack, WhatsApp, Gmail & Google Workspace
  • Webhooks, ETL jobs and rate-limit-safe sync
05

RAG & Knowledge Systems

Answers with receipts. Retrieval over your documents, wikis and drives, wired into whichever model you want, with citations your users can check.

shipped: LawGPT · archestra-ai connectors · Chatter AI

  • Embeddings and vector store design
  • Chunking, retrieval tuning and reranking
  • Citation output and grounding checks
  • Provider-agnostic model layer (Claude / OpenAI / Gemini)
06

MCP Servers & Agentic Workflows

Custom Model Context Protocol servers and multi-agent flows that give AI real, permissioned access to your systems — with human approval gates where it matters.

shipped: AI DevOps Agent · slack-claude · QuickDocs · archestra-ai

  • Custom MCP servers for your APIs and data
  • Planner / executor and supervisor agents (LangGraph)
  • Tool permissions, audit trails, human-in-the-loop
  • Claude Code / Cursor workflow setup for your team
07

LLMOps, Evals & Guardrails

The difference between a pilot and production. Tracing, eval suites, prompt versioning, and the cost and latency work that keeps the AI feature affordable.

shipped: Production AI at Loadshare Networks · archestra-ai · Chatter AI

  • Tracing and observability (LangSmith)
  • Eval suites and regression tests for prompts
  • Cost, latency and token optimisation
  • PII handling and prompt-injection defence
08

Workflow & GTM Automation

The boring, expensive manual work — gone. n8n pipelines, internal bots, lead enrichment and outbound that run quietly in the background.

shipped: Lead Enrichment Tool · Cold Outreach Engine · Cloudflare Tunnel + n8n @ Data Alt Dynamics

  • n8n and custom pipelines across your stack
  • Lead scraping & AI enrichment
  • Personalised outbound with caps & suppression
  • Reporting and spreadsheet automation
09

Physical AI & Robotics Integration

AI that touches hardware. ROS2 control stacks, robot arms, IoT sensor networks and live factory dashboards — from firmware all the way up to the browser.

shipped: BCN3D Moveo · SCARA Robot · Sensor Dash · UK Design 6450987

  • ROS2 control stacks and embedded firmware
  • Robot arm integration (6-DOF, SCARA)
  • IoT sensor networks and OEE dashboards
  • Voice and LLM control of real machines

How it works

01

Scope call

30 minutes at cal.com/ayuugoyal. You describe the problem; I tell you straight whether AI is the right answer and what it would take.

02

Embed & discover

I get access to your stack and time with the people doing the work. The real requirements live with users, not in the brief.

03

Ship to prod

A working pilot early, in your environment, then hardened for production: auth, monitoring, evals, fallbacks.

04

Hand over

Documented and handed to your team — or kept on a retainer if you would rather I keep running it.

Deployment sprint
One well-defined use case, taken from zero to production over a fixed few weeks.
Pilot to production
You already have a demo that works on someone's laptop. I make it survive real users.
Embedded retainer
Part-time FDE on your team, shipping and maintaining AI features month to month.

Questions I actually get

What is a forward deployed engineer?

+

A forward deployed engineer (FDE) is a software engineer who embeds directly with a customer's team to make a product or technology work in their real environment. Instead of handing over a spec, an FDE learns the workflow from end users, writes the integrations, and ships to production. For AI, that means taking a model from demo to something your team uses every day.

Can I hire Ayush Goyal as a freelance forward deployed engineer?

+

Yes. Ayush takes a small number of freelance forward-deployed engagements, fully remote and worldwide. Book a free 30-minute scope call at cal.com/ayuugoyal or email ayushgoyal8178@gmail.com with what you are trying to ship.

How is a forward deployed engineer different from an AI consultant?

+

A consultant typically leaves you with recommendations. A forward deployed engineer leaves you with running software: code in your repositories, integrations with your systems, and monitoring in production. The advice is a side effect of doing the work.

How do you freelance while working full-time as an AI Engineer?

+

Ayush works full-time as an AI (Harness) Engineer at TAP Innovations and takes a limited number of scoped freelance engagements on top. Timelines and availability are agreed up front on the scope call, so there are no surprises mid-build.

Can you work inside our existing stack and security rules?

+

That is the point of forward-deployed work. Ayush builds provider-agnostic, with Claude, OpenAI or Gemini selected by config, and has shipped across Next.js, FastAPI, Flask, Express, Postgres, MongoDB, Docker, AWS and Azure. Ayush works within your access controls rather than around them.

What does a typical engagement look like?

+

It starts with a scope call, then an embed-and-discover phase with your users, a working pilot in your environment early, a production rollout with monitoring and evals, and a clean handover. Shapes range from a fixed deployment sprint to an ongoing embedded retainer.

Who is Ayush Goyal?

+

Ayush Goyal is a Forward Deployed Engineer and AI Engineer based in India and working remotely worldwide. Ayush takes freelance forward-deployed AI engagements, works full-time as an AI (Harness) Engineer at TAP Innovations, builds robots as side projects, holds UK Design Registration 6450987 for a pneumatic robotic gripper, and has earned $400 in open-source bounties from archestra-ai.

What does Ayush Goyal build?

+

Production AI systems — including a WhatsApp machine-maintenance ticketing bot live on a factory floor and SiteOS, a WhatsApp-to-dashboard platform for an architecture studio — plus AI agents and chatbots, WhatsApp bots, bilingual Hindi and English voice agents, RAG and knowledge systems, MCP servers, enterprise integrations and workflow automation — plus robots and Physical AI: ROS2 robot arms, a SCARA, voice-controlled ArduPilot vehicles and IoT factory sensor servers.

Can you build voice agents that speak Hindi and English?

+

Yes. Bilingual voice agents that handle inbound and outbound calls, qualify leads and book appointments, with telephony integration and transcripts synced back to your CRM.

Can you build WhatsApp bots and AI agents?

+

Yes. Two WhatsApp systems are live in production from freelance FDE work: a machine-maintenance ticketing bot on WhatsApp Flows for a factory, and SiteOS, which turns site engineers' WhatsApp photos and notes into tickets and daily reports. Personal builds include a FastAPI + Gemini assistant for National Building Code questions that runs on a small VPS.

Can you integrate AI with the tools we already use?

+

That is most of the job: Slack, WhatsApp, Gmail and Google Workspace, Notion, SharePoint, OneDrive and Drive, CRMs, calendars and internal databases — with OAuth, retries, rate limiting and error handling built in.

What robots has Ayush Goyal built?

+

A full control system for the BCN3D Moveo 6-DOF robot arm (ROS2, Arduino, Next.js), a 4-axis SCARA robot controlled from the browser, ArduPilot Assistant for natural-language and voice control of ArduPilot vehicles over MAVLink, a Raspberry Pi IoT factory sensor server, and a 3D-printed pneumatic four-finger gripper registered as UK Design 6450987.

Is robotics your job or a side project?

+

Side projects. Ayush's day job is AI engineering and the freelance work is forward-deployed AI engineering; robots are what Ayush builds after hours, often through RnDynamos Labs. That hardware work is available on client projects too.

What robotics stack do you use?

+

ROS2 Humble and ROSBridge, Arduino and embedded C++, Raspberry Pi, stepper motion control, MAVLink and ArduPilot, WebSockets for live control, Next.js for web control panels and 3D visualisation, and FDM and resin 3D printing for parts.

Can AI and LLMs control real robots?

+

Yes, with a safety layer in between. ArduPilot Assistant turns spoken or typed commands into MAVLink through an LLM orchestration layer, and the same pattern applies to robot arms: the model proposes, a deterministic controller validates and executes.

got something that needs a brain?

book a scope callsee the work