Sheikh Shihab HossainI build LLM systems that can be measured.

Everything, in order

Every one of the thirteen, and who it was built for. Role spans are exact; the month on each item is an estimate.

Where I am

I'm a Full-Stack Engineer at Sharetasking Inc. The job is SiloCRM, the company's AI-powered CRM platform, which I build alongside a team of 5+ engineers. I work across the whole stack there: Next.js and TypeScript on the frontend, Fastify and Prisma on the backend, PostgreSQL with row-level security at the data layer.

Most of my recent work at Sharetasking has been LLM application engineering on SiloPilot, SiloCRM's multi-channel AI agent. I built its core integrations: a real-time streaming voice agent on ElevenLabs ConvAI (WebSocket), AI-generated SMS replies on Twilio, RAG pipelines with per-tenant knowledge-base sync, tool-use for call transfer and data collection, and the agents management UI. That work shaped my opinions on prompt engineering, context architecture, and what production LLM systems actually need beyond the model call.

Also at Sharetasking, I designed and built SiloCRM's visual workflow automation engine end-to-end: an n8n-style canvas with 40+ triggers, 20+ actions, and a custom Fastify execution engine. Both SiloPilot and the workflow engine are company products, not side projects.

On my own time I built two things. Zaxvio CRM is a multi-tenant field-service SaaS for HVAC contractors, solo and end-to-end. EvalLab is an open-source LLM evaluation harness (Python/FastAPI + Next.js) with LLM-as-judge scoring, failure clustering, and judge calibration; I wrote it because eval discipline is usually too thin in shops shipping LLM features, and I wanted a tool that fixed that for me.

ROLE

Full-Stack & LLM Engineer

COMPANY

Sharetasking Inc.

BASED

Khulna, Bangladesh

EXPERIENCE

2+ years

AVAILABILITY

Open to senior full-stack roles

RIGHT NOW

Building Multi-judge ensembles and webhook-triggered eval runs for EvalLab. Last shipped EvalLab, an open-source LLM evaluation harness (Python/FastAPI + Next.js).

What I actually work in

Sorted by how far it has been pushed, not by how impressive the logo is. A tag cloud claims everything equally; this says which things have carried real traffic.

Roles, in order

1. 02Aug 2025 - Present

   Full-Stack Engineer

   Sharetasking Inc.

   • Contributing to SiloCRM (silocrm.com), a multi-tenant CRM monorepo (Next.js, Fastify, Prisma, PostgreSQL with row-level security), alongside a 5+ engineer team.
   • Built core LLM integrations for SiloPilot, the platform's per-tenant multi-channel AI agent: real-time streaming voice on ElevenLabs ConvAI (WebSocket), AI-generated SMS on Twilio, RAG pipelines with per-tenant knowledge-base sync, tool-use for call transfer / data collection, conversation logging substrate for observability, and the AI agents management UI.
   • Designed and built end-to-end the visual workflow automation engine, React Flow canvas, 40+ trigger types, 20+ actions, custom Fastify execution engine with breadth-first graph traversal and per-node execution logs.
   • Daily Claude Code and Cursor user,directing agents, curating context, and reviewing outputs across the monorepo.

   LLMRAGFastifyNext.jsTypeScriptPrismaPostgreSQLElevenLabsTwilioReact Flow

2. 01Aug 2024 - July 2025

   Software Engineer

   Ternary Solutions Inc.

   • Built and maintained multi-tenant SaaS products with Next.js, TailwindCSS, and Django, with role-based access control across admin, client, and team-lead surfaces.
   • Shipped responsive, component-driven dashboards and reusable UI primitives; integrated Django REST APIs and contributed backend changes when needed.
   • Contributed to Counterfoil (frontend of a modern SaaS platform) and Alley Analytix (role-based dashboards + marketing site with Payload CMS).

   Next.jsDjangoPostgreSQLTailwindCSSFramer Motion

Selected work

Four of these are production systems with real users behind them. Where the source is closed, it says so.

Open-source LLM evaluation pipeline. A personal project, designed and built end-to-end.

Live siteSource

EvalLab

An open-source evaluation harness for LLM products: define test sets, define agents, run them against an LLM-as-judge, compare prompt versions with reproducible scored results. The idea: ship prompts with numbers, not vibes. Built after working on production LLM agents and noticing how thin the eval discipline usually is.

Implements LLM-as-judge scoring across four dimensions (accuracy, completeness, tone, safety), failure clustering that groups low-scoring cases by failure mode, and judge calibration with Cohen's κ and a 5×5 confusion matrix between human and LLM judge scores. Runner uses a token-bucket rate limiter, exponential-backoff retries, per-case error isolation, and immutable agent/prompt versioning so historical scores stay reproducible.

PythonFastAPISQLAlchemy 2.0PostgreSQLNext.jsTypeScriptTailwindCSSshadcn/uiGroqLLM-as-judgeRAG-adjacentRenderVercelNeon

Multi-tenant field-service SaaS for HVAC contractors. Built solo, end-to-end.

Live siteSource

Zaxvio CRM

A field-service platform for solo HVAC contractors and 1-3 person teams, replacing the phone-and-paper workflow with scheduling, quoting, invoicing and customer history in one place. Turborepo monorepo: a Fastify API and a Next.js 14 App Router frontend over Supabase Postgres with Drizzle ORM.

24 API modules covering jobs, quotes, invoices, customers, equipment, maintenance contracts, online bookings, checklists, a product catalog and configurable sales pipelines. Multi-tenant throughout, with a superadmin surface for tenants, affiliates, analytics and support.

It also carries its own workflow automation engine, with a graph schema, run history and integration events, plus Better Auth for org-scoped access, Lemon Squeezy billing, Mapbox routing, generated PDF invoices, and Supabase Realtime. 186 commits, mine.

TypeScriptNext.js 14FastifyTurborepoDrizzle ORMSupabasePostgreSQLBetter AuthLemon SqueezyMapbox GLResendVitestPlaywright

A feature of SiloCRM. Designed and built end-to-end, solo.

Live siteSOURCE CLOSED

Workflow Automation Engine

A visual no-code workflow builder I designed and built end-to-end for SiloCRM. Users drag triggers, conditions, and actions onto a React Flow canvas to automate CRM processes like 'when a new lead comes in, tag it, wait 5 minutes, send an SMS, create a task.'

40+ trigger types (lead lifecycle, calls, SMS, email events, Facebook leads, scheduled, external webhooks with HMAC). 20+ actions across CRM mutations, messaging, HTTP requests, Google Sheets, Slack, GA4 / Google Ads / Meta CAPI.

Custom Fastify execution engine with breadth-first graph traversal, per-node execution logs, a replayable read-only viewer, and a shared variable schema with {{contact.first_name}}-style templating resolved at runtime.

The example above, executed breadth-first, one depth at a time, with a log line per node. That is how the engine actually runs a graph.

React FlowZustandFastifyPrismaPostgreSQLRedisNext.js

A multi-channel LLM agent in SiloCRM. I built the core integrations, on a team of 5+.

Live siteSOURCE CLOSED

SiloPilot

Per-tenant LLM-powered agent for SiloCRM that handles inbound and outbound voice, SMS, and email, used for lead qualification, appointment booking, data collection, and call routing. Each agent is bound to a tenant-owned Twilio number.

I built the ElevenLabs ConvAI integration (real-time streaming LLM responses over WebSocket), the RAG layer (per-tenant knowledge-base sync against ElevenLabs' document store, with chunking and tool definitions), the Twilio SMS layer with AI-generated replies, the conversation logging substrate for observability, and the AI agents management UI.

LLMRAGElevenLabs ConvAITwilioFastifyNext.jsReal-time WebSocket

A Chrome new tab page. Open source, 12 stars.

Source

TabDeck

A feature-packed new tab page for Chrome with bookmarks, widgets, and a focus-friendly layout. My most-starred public project.

JavaScriptChrome Extension

What people who shipped with me say

Straight answers

Who is Sheikh Shihab Hossain?

Sheikh Shihab Hossain is a Full-Stack and LLM Engineer based in Khulna, Bangladesh. He currently builds LLM-powered agents and workflow automation at Sharetasking Inc. on the SiloCRM platform, and is the creator of EvalLab, an open-source LLM evaluation pipeline.

What does Sheikh Shihab Hossain work on?

He ships production multi-tenant SaaS and LLM-powered conversational agents: real-time streaming voice, SMS bots, RAG pipelines, and visual workflow automation engines. Outside work he built Zaxvio CRM, a multi-tenant field-service SaaS for HVAC contractors, solo and end-to-end. His stack is Next.js, Fastify, FastAPI, Python, TypeScript, and PostgreSQL.

What is EvalLab?

EvalLab is an open-source LLM evaluation pipeline created by Sheikh Shihab Hossain. It lets teams define test sets and agents, run them against an LLM-as-judge across accuracy, completeness, tone, and safety, and compare prompt versions with reproducible scored results. It also includes failure clustering and judge calibration via Cohen's κ.

Is Sheikh Shihab Hossain available for hire?

Yes. He is open to senior full-stack roles, particularly involving LLM application engineering. The fastest way to reach him is sheikhshihabhossain2020@gmail.com.

Where is Sheikh Shihab Hossain based?

Khulna, Bangladesh. He works remotely with USA-based companies and is comfortable with UK-aligned hours and distributed engineering teams.

What technologies does Sheikh Shihab Hossain specialize in?

Frontend: Next.js, React, TypeScript, TailwindCSS, shadcn/ui, Framer Motion, React Flow. Backend: Fastify, Node.js, FastAPI, Django, Prisma, SQLAlchemy 2.0. Data: PostgreSQL with row-level security, Redis, Neon. LLM: ElevenLabs ConvAI, Twilio, Anthropic / OpenAI / Groq APIs, RAG, prompt engineering, LLM-as-judge evaluation. He uses Claude Code and Cursor daily.

Open to senior full-stack roles, particularly LLM work.

Email is the fastest way to reach me. I work remotely with US-based teams and I’m comfortable with distributed hours.

• GITHUBgithub.com/iamshihab2020
• LINKEDINlinkedin.com/in/sheikh-shihab-hossain
• Xx.com/theshihab2022
• CURRICULUM VITAEPDF, 2 pages
• PHONE+880 1776 600718
• PHONE, ALT+880 1954 114410