Starter
AI feature added to an existing product. 4–6 weeks.
On request
Final number lands in the discovery call.
- 1 AI feature, integrated into your stack
- Eval harness + cost model
- Production deployment
- 30-day post-launch support
Services /AI Solutions
AI Solutions.
Custom AI products, copilots, and automation built by a founder shipping 10 AI startups in parallel.
OpenAIAnthropicCustom ModelsVoice AI
FIELD · AI
EST. 2011
10
AI STARTUPS SHIPPING
Live across Voxly, NOBBYO, Pannakhata
The problem
AI integration is hard. Most of it isn't AI.
Founders come to me with an 'AI feature' they need shipped. 80% of the work is the unsexy part: data plumbing, evaluation harnesses, fallback behavior, cost monitoring, observability. The model is the easy part. The product around it is where AI projects live or die.
Your team can prompt an LLM but can't ship one to production safely.
You're spending more on tokens than on the engineers running them.
Hallucinations break demos — and you don't have a system to catch them.
You shipped an AI feature six months ago. Nobody knows if it's working.
The solution
AI products built by someone shipping 10 in parallel.
I run 10 AI startups concurrently — Voxly, NOBBYO, Pannakhata, VoiceBridge, and six more. Every constraint your AI build will hit, I've hit this quarter. The playbook is shipped, not theoretical.
Production-grade integration in 6–10 weeks, not 6–10 months.
Evaluation harness on day one — you know if accuracy regresses.
Cost ceiling enforced via provider routing — no surprise OpenAI bills.
Observability built in — drift, hallucinations, and usage tracked from launch.
What's included
Everything in scope.
What you get in every AI Solutions engagement, regardless of tier.
- Discovery + AI feasibility audit on your problem
- Custom integrations with OpenAI, Anthropic, or open-source models
- RAG pipelines, fine-tuning, and eval harnesses
- Automation workflows that move work, not just data
- Cost-per-conversation modelling before I ship
- Production deployment + 30 days post-launch support
Process
How AI Solutions engagements run.
The same four-step rhythm used across every service, tuned for this specific work.
- Step 1
Discover
Two calls and a written brief. I separate the problem you have from the AI demo you saw last week. Most teams arrive with a solution looking for a problem — together we leave Discover with a problem that's specific enough to scope.
- Step 2
Design
Eval harness first, UI second. I define what 'good' looks like in measurable terms — accuracy, latency, cost-per-call — before any model selection. The UI gets designed against the eval, not the other way around.
- Step 3
Build
Two-week sprints with weekly working demos. You see the actual product behaving against your data before week 3. I pick models for your eval, not for hype, and swap them as cheaper or smarter ones land.
- Step 4
Ship
Deploy to your stack with monitoring on key metrics from day one. 30 days of post-launch support included so the first week of real-user data doesn't catch you alone.
Tools · tech
Stack I ship with.
- OP
OpenAI
- AN
Anthropic
- VA
Vercel AI SDK
- LA
LangSmith
- PI
Pinecone
- NJ
Next.js
Pricing
Three tiers for AI projects.
Final numbers land after a discovery call so the tier matches actual scope. Tier shapes are stable.
Pro
AI MVP — copilot, agent, or end-to-end product. 8–12 weeks.
On request
Final number lands in the discovery call.
- Multi-feature product or copilot
- RAG + custom fine-tuning where it pays off
- Eval harness + ongoing benchmark dashboard
- Two production deployments + monitoring
- 60-day post-launch support
Enterprise
Strategic AI engagement with embedded retainer. Quarterly cycles.
On request
Final number lands in the discovery call.
- Embedded AI lead working with your team
- Roadmap + quarterly velocity
- Custom model fine-tuning where the data warrants
- Compliance + data-handling review
- Always-on support
Live proof
AI Solutions work, shipping right now.
Three of my own ventures using exactly the AI project playbook I'd run for you. Live receipts, not Phase-10 mockups.
ORBIX
Stage · LIVEStarted · EST 2024Users · 500+ BUSINESS USERS
AI-powered Business OS, live across BD, UK, and Luxembourg with 500+ business users.
Voxly
Stage · BETAStarted · MAR 2025Users · 240 BETA TESTERS
Real-time voice-to-text for Bangla and other under-served South Asian languages.
NOBBYO
Stage · BETAStarted · DEC 2024Users · 400 ACTIVE SELLERS
AI knowledge ops — copilot for non-technical operators.
Meet the team
Who's actually on your project.
Md. Ersaduzzaman (Antor)
Founder, lead architect
Currently shipping 10 AI startups. I write production code, set evaluation criteria, and review every release before it ships to your users.
Joya
Managing Director, NextBangla
Operational backbone. Project schedules, contracts, invoicing, communication cadence. You'll hear from her at least as often as me.
[Senior AI Engineer — name on contract]
Implementation engineer
Senior engineer assigned per project based on stack. Names disclosed once an NDA is in place.
Why hire me for this
Three reasons specific to AI projects.
I'm shipping 10 AI startups right now
Voxly, NOBBYO, Pannakhata, VoiceBridge — when I build AI into your product, the same patterns I'm using shipped in mine last week. You're not the test case.
Eval-first, not demo-first
Most AI projects look great in a demo and fall over in week 2 of real traffic. I define 'good' with numbers before I pick a model. That's why my products survive past the launch tweet.
Cost-per-call is part of the spec
An AI feature that costs $4 per session can sink a unit economics model. I model cost before I ship, and I revisit it every time a cheaper model lands — which is roughly every six weeks now.
FAQ
AI Solutions questions.
Start your AI project
Let's talk about your AI project.
If the scope above looks like what you need, the next step is a 15-minute discovery call. No pitch — just a conversation about what you're building.