Skip to contentAbu Bakar Siddik

Work with me

Fine-tuning, private LLM deployments, and AI agents for production.

I lead AI engineering full-time at a legal-AI company, and I take on a small amount of outside work. Most of it is one of three things: fine-tuning an open model on a team's own data, running models where the data can't leave the building, or fixing an agent or RAG system that works in a demo but not for real users.

If something here sounds like what you're dealing with, send me a note. The first call is free.

What I can help with

  • Fine-tuning open models (LoRA, QLoRA, SFT) on your data, with evaluations that show whether it actually helped
  • Private deployment: open or fine-tuned models on your own GPUs, in your VPC, on-premise or fully air-gapped
  • Zero data retention: setting up OpenAI, Anthropic, Vertex AI, Azure OpenAI, Bedrock and others so prompts and outputs are not kept
  • Agents that call real tools, recover from failures and finish multi-step work in production
  • RAG and search over your documents and databases, with citations that stay grounded
  • Evaluation and hardening of an agent you already have, including prompt-injection risks

Ways to work together

First callFree30 minutes
We talk through your problem. Afterwards I send a short written note on what I would build and how. It is yours to keep either way.
Setup review$250fixed price · written report in 3 working days
A review of an existing zero-data-retention, private LLM, agent or RAG setup: what works, what is exposed or fragile, and exactly what to change.Included at no extra cost
  • A 30-minute call to walk through the findings
  • One follow-up question round by email
Prototype$1,000fixed price · about 2 weeks
A working prototype of one agent, RAG workflow or fine-tuned model on your data, running in your environment, so you can see it work before committing to more.Included at no extra cost
  • Scoping call and a written plan
  • An evaluation set built from your real cases, so you can see how well it works on your data
  • Code in your repository with setup docs
  • A recorded walkthrough for your team
  • Two weeks of fixes after hand-over
Production buildfrom $6,000fixed price · 4–8 weeks
The full system, taken from prototype to production: architecture, fine-tuning if it helps, deployment to your cloud, VPC or on-premise servers, and monitoring.Included at no extra cost
  • Zero-data-retention and privacy review of every LLM provider you use
  • Evaluation suite that runs on every change
  • Hand-over sessions with your engineers
  • 30 days of support after launch
Advisory$2,000 / monthabout 8 hours a week · cancel any time
Regular time with your team on architecture, code review and model choices.Included at no extra cost
  • The first week is free
  • Async questions answered within a working day

Introductory rates while I take on my first outside clients. Every project is a fixed price agreed in writing before work starts; no hourly billing.

Where I've done this before

AskTuring.ai
Led the Core RAG & AI team and owned the platform architecture while usage grew from 100 to 10,000+ users; rebuilt the database layer to hold 1,500 concurrent AI requests. How we scaled it.
Sazim Tech
Built and shipped private LLM deployments for enterprise clients, including fully air-gapped installations with self-hosted models and no internet connection, behind one provider layer for OpenAI, Anthropic and local models.
Fine-tuning
A LoRA for Qwen3.5-2B that picks the right tools 84.6% of the time against 41.5% for the base model, and a QLoRA fine-tune of Qwen3.5-4B whose tool calls are valid 97.5% of the time. More on Hugging Face.
In the open
Golem, my Go agent framework, and guides to zero data retention across LLM providers.

Common questions

How much does it cost to hire an AI engineer?

With me: the first call is free, a setup review is $250, a two-week prototype is $1,000, a production build starts at $6,000, and advisory is $2,000 a month with the first week free. Each project is a fixed price agreed in writing before work starts.

How does working together usually start?

With an email about the problem, your stack and your timeline. We have a free 30-minute call, I send you a short written plan, and if you want to go ahead we agree scope and a fixed price in writing.

What kinds of AI systems do you build?

Fine-tuned open models, private and air-gapped LLM deployments, AI agents and multi-agent workflows, RAG and search over company data, and MCP integrations. Public examples include fine-tuned models on Hugging Face, the Go agent framework Golem and the self-hosted AI platform Herbie.

Can you fine-tune a model on our data and run it privately?

Yes. I fine-tune open models such as Qwen, Gemma and Llama with LoRA or QLoRA, measure them against the base model on held-out data, and deploy them on your own GPUs, in your VPC, or fully air-gapped, so your data never leaves your environment.

Can the AI run inside our own infrastructure?

Yes. VPC, on-premise and fully air-gapped deployments with self-hosted models are all options, so sensitive data can stay in your environment.

Can you work inside our existing codebase?

Yes. I use Python, TypeScript, Go and Rust day to day, and many projects start as an audit or a small integration into what you already run.

Who owns the code?

You do. Work happens in your repository where possible, ownership transfers on payment, and NDAs are fine.

Can you help with zero data retention?

Yes. I can review a ZDR or private LLM setup against each provider's current terms and tell you what is covered, what is still retained, and what to change. My zero-data-retention guides cover the background.

Where are you based?

Rajshahi, Bangladesh (UTC+6), working remotely. That overlaps European business hours and US mornings or evenings.

Get in touch

Tell me what you're working on and where it's stuck, and we'll set up the free first call. I reply within a working day, even if I'm not the right person for it.

Email meabubakar1808031@gmail.com

Prefer to talk? Book a 30-minute call.