Available for Forward Deployed / AI Engineer roles — remote worldwide

I build AI systems that stay in production.

Forward Deployed AI Engineer. 50+ client deployments across 8 industries — legal, finance, manufacturing, retail, agriculture, healthcare. Multi-tenant RAG, LangGraph multi-agent orchestration, and self-hosted LLM inference on a 2× RTX 5090 cluster. Not demos. Production systems businesses run on every day.

50+Production deployments
8Industries shipped
50+Concurrent AI agents
64GBSelf-hosted GPU VRAM
About

I walk into a business that can't fully explain its problem, figure out what's actually broken, and ship an AI system that stays in production. I've done this 50+ times across legal, finance, manufacturing, retail, agriculture, healthcare, e-commerce, and education.

I didn't study computer science. I studied commerce (M.Com, Kota University) and learned to build by building — for real clients, with real deadlines, real money on the line. That's not a weakness. It's why my deployments get adopted: I read the workflow and the P&L before I write a spec.

Right now I lead a 20-person team at Gosotek Technology, architecting production AI — multi-tenant RAG, multi-agent orchestration, and a self-hosted 2× RTX 5090 inference cluster running quantized Qwen-72B via vLLM. I own the metal, the evals, and the outcomes.

Forward Deployed Engineering

Discovery to deployment, end to end. Scoping, POC, pilot, production, iteration — 50+ engagements shipped.

AI / LLM Systems

RAG, multi-agent orchestration, QLoRA fine-tuning, self-hosted vLLM inference. Below the API layer.

Revenue Operations

CRM/ERP architecture, lead routing, automation, AI sales intelligence. Revenue systems for 50+ companies.

Case Studies

Production deployments, with the numbers.

Each of these is live, used daily, and shipped end-to-end — discovery call to production to iteration.

Legal LLM — On-Premise Multi-Tenant RAG

Multinational Law Firm

Privacy-first legal AI on the client's own 4× RTX 5090 cluster (1TB NVMe). Attorneys tag the AI in WhatsApp for citation-grounded answers; it's wired into a locally-hosted Frappe ERP so it can take operational actions. Per-company data isolation across 6–8 tenants, with zero cross-tenant leakage. A 340-case golden-set eval gates every deploy.

RAGChromaDBHybrid SearchCross-encoder RerankingWhatsApp APIFrappe ERPOn-Premise
15–30hSaved / week / attorney
+50%Case capacity
71→89%Citation precision
~2–3%Hallucination rate

OpenSalesAI — LangGraph Multi-Agent Sales Platform

FMCG / CPG Distributor

A LangGraph StateGraph supervisor detects intent and routes to 5 specialist agents — order, coaching, analytics, collection, promotion — over a Qdrant + LangChain RAG pipeline with a 3-tier LLM fallback chain. TypeScript dashboard + Python FastAPI service, streaming responses, per-agent context budgeting. Deployed across 50–500 retail outlets.

LangGraphMulti-AgentQdrantFastAPIPydanticStreamingMLflow
5Specialist agents
3-tierLLM fallback chain
50–500Retail outlets

MyKiranaBuddy — B2B Wholesale Commerce

Retail / FMCG

B2B wholesale commerce platform with a field sales app, route optimization, and Frappe ERP integration. I owned the full revenue engine — acquisition, conversion, retention, analytics — growing it to 5,000+ users at ₹50L revenue with 70–80% month-over-month growth.

FlutterFrappe ERPRoute OptimizationField Sales App
5,000+Users
₹50LRevenue built
-40%Logistics cost
+60%Upsell lift

Finance Client — Voice AI Onboarding & Fraud Prevention

Financial Services

Voice-to-voice onboarding with GPS scoring for Tier 3–4 users. A Grok API voice-auditing agent calls customers post-onboarding with 2-factor auth to verify leads. Cut fake profiles and scaled loan disbursement 3×.

Voice-to-VoiceGrok APIGPS Scoring2FA
2–3 minOnboarding (was 15–20)
-80%Fraud
Loan volume

Garment Manufacturing — Full AI Digital Transformation

Clothing Manufacturer

Complete digital transformation for a non-tech company: WhatsApp + in-app AI chatbot, 11Labs voice agent, open-source Frappe ERP with a natural-language AI assistant, and a Claude Code-powered staff training system. Added a Gemini transcription layer that cut Indian-language voice errors dramatically.

Frappe ERP11Labs VoiceClaude CodeMulti-Language
30–40% → 2–3%Transcription error rate

ReVideoz — Social Media Intelligence

Marketing Agency

Multi-platform scraping, transcription, and psychological content analysis across Instagram, X, TikTok, YouTube, Reddit, Facebook. A parallel scraping engine runs 8–13 concurrent agents with multi-provider key rotation, plus multimodal Gemini analysis and a content-recreation pipeline.

Gemini APIParallel AgentsApifyHiggsfield/Kling
95%Content similarity
8–13Parallel agents
6Platforms
Open Source

Code I've shipped publicly.

Open-source repositories — agent tooling, multi-agent harness, sales AI, automation libraries. All on github.com/Akshit1018.

Client Work — Live Sites

28 production websites, shipped.

A selection of live client websites and platforms I've built or led. The rest live behind the PIN-gated Projects page.

Software, Side Projects & Mobile

Things I've built and shipped.

Apps, prototypes, and experiments — deployed on Vercel and custom domains.

Latest Projects — Confidential

Some work stays under NDA.

The latest deployments — including unreleased client work — live behind a PIN. Message me on WhatsApp to get access.

Want to see everything I'm building right now?

Recent and confidential client deployments are gated behind a PIN. Ping me on WhatsApp and I'll send it over — happy to walk you through the architecture and the evals.

Writing

Field notes from production AI.

What I learn shipping RAG, multi-agent systems, and self-hosted inference — written down so I don't make the same mistake twice.

View all 18 posts →

Client References

What clients say.

Client references are available on request — most of my work is under NDA, so I share verbatim quotes and references directly over a call or WhatsApp.

"

The legal AI platform cut our research time by more than half. It's now part of how we work every single day.

Managing Partner · mid-size law firm (reference on request)
"

He came in, understood our sales workflow in a week, and shipped something the field team actually uses. Rare.

Operations Lead · FMCG distributor (reference on request)
"

For a non-tech company, he made AI feel boring in the best way — it just works, and our staff use it without training.

Director · garment manufacturer (reference on request)

Request references on WhatsApp →

Let's build something that ships.

Available for Forward Deployed Engineer / AI Engineer roles, and selective client work — remote worldwide.