jaydeep raijada

Work with me

I help teams and businesses build and ship AI, from custom model fine-tuning to RAG, agents, and automation. I'm an applied AI engineer who trains and deploys models in the open; you get the depth without the agency overhead.

Currently taking on a few projects. Reach me at j.raijada25@gmail.com.

What I build

Model Fine-Tuning & Evaluation

For teams that have outgrown prompting. I domain-adapt or post-train open models so they're faster, cheaper, and better at your specific task, and I build the eval harness to prove it.

  • —SFT / LoRA / QLoRA / continued pre-training
  • —RL post-training (GRPO, reward design)
  • —Eval suite + benchmark vs. your current setup
  • —Deployable checkpoint + handoff

RAG & Knowledge Systems

Turn your documents, catalog, or knowledge base into accurate, grounded answers with citations, so you can trust the output.

  • —Retrieval pipeline over your data
  • —Grounded, cited responses
  • —Evaluation for accuracy & hallucination
  • —API or chat interface

AI Agents & Automation

Tool-using agents and workflow automation that take real work off your team. For businesses: WhatsApp & voice assistants that answer customers, capture leads, and book appointments 24/7.

  • —WhatsApp / voice assistants (multilingual)
  • —Lead capture, FAQs & booking automation
  • —Tool-using agents for internal workflows
  • —Setup + ongoing maintenance

AI Audit & Advisory

A focused sprint to assess an existing LLM system or map where AI can actually help your business, ending in a concrete, prioritized roadmap you can act on.

  • —System / opportunity review
  • —Eval harness for what you already run
  • —Prioritized roadmap + cost estimate
  • —1–2 week turnaround
How it works
  1. 01
    Intro callA free 20-minute call to understand the problem and whether I'm the right fit.
  2. 02
    Fixed-scope proposalClear deliverables, timeline, and price. No open-ended hourly billing.
  3. 03
    BuildI build it, with progress you can see along the way.
  4. 04
    Handoff & supportDeployed, documented, and maintained, with ongoing support where it makes sense.
Why me

I don't just integrate APIs. I train models from scratch and ship them publicly. I trained a 1.5B RL safety monitor to 0.893 AUROC (Top 100 at the Hugging Face × Meta OpenEnv Hackathon), domain-adapted a small model for a 20% perplexity drop, and built a full LLM training stack by hand. I've also shipped this end-to-end: TenderIQ, a production RAG system with cited verdicts, OCR fallbacks, and a human-review queue, and an LLM-based customer-feedback pipeline running in production at a Fortune 50 retailer. That depth is what you get on your project.

Selected workGitHubHugging Face
Common questions
Do you work remotely?
Yes. I'm based in Bangalore (IST) and work with clients anywhere, mostly async with scheduled overlap for calls.
Who owns the work?
You do. Code, model weights, data, and documentation are all handed over at the end, and I'm happy to sign an NDA before we talk specifics.
What size of project do you take on?
Anything from a 1–2 week audit or eval sprint to a multi-month build. If a project needs more hands than mine, I'll say so up front.
What should my first email include?
What you're trying to do, what data or system you already have, and your rough timeline. Two or three sentences is enough — I'll come back with questions or a call slot.
Get started

Tell me what you're trying to do and I'll tell you honestly whether AI is the right tool and how I'd approach it. Email j.raijada25@gmail.com or reach me on X and LinkedIn.