Skip to content
SP Sarath Panha

Freelance · AI & Software Architect · Booking from [month]

I design and ship AI systems that hold up in production.

From LLM and RAG features to the cloud architecture behind them, I help startups and product teams build AI that is reliable, secure and affordable to run.

Book a discovery call See case studies
A typical RAG feature The user app calls an API, which retrieves context from a vector store and calls the LLM, with evals and monitoring alongside. Product app AI service guardrails, cache LLM API Vector store Evals Tracing, cost and quality dashboards top-k context usage events
fig. 1 · the shape of a typical RAG feature I deliver
Years building software
[0]
Production systems
[0]
AI features shipped
[0]
Users served
[0k]

Services

Three ways to work together

Compare services →
1 to 2 weeks

AI architecture review

An outside look at your existing or planned AI system: risks, cost, reliability, and a target architecture with a prioritised plan.

Learn more → 4 to 10 weeks

AI feature build

A RAG, agent or LLM feature designed, built and shipped to production, with evals, guardrails and monitoring in place.

Learn more → Monthly retainer

Fractional architect

Ongoing architecture leadership for your team: design reviews, decision records, code reviews and mentoring.

Learn more →

Selected work

Case studies

All work →

Sample projects below show the format. Replace them with your own.

AI / LLM · RAGSample

RAG support assistant for a SaaS help desk

Grounded answers over 12k help articles, with citations and an evaluation suite.

Example outcome: resolution time down 40%

Cloud · BackendSampleTemplate only

Multi-tenant API platform on serverless

Re-architected a monolith into tenant-isolated services with per-tenant cost tracking.

Example outcome: infra cost down 35%

AI / LLM · DataSampleTemplate only

Document extraction pipeline with LLMs

Structured data from invoices and contracts, with human review only where confidence is low.

Example outcome: 80% of documents fully automated

How an engagement works

From first call to handover

  1. Step 1

    Discovery call

    30 minutes, free. We talk through the problem, constraints and what success looks like.

  2. Step 2

    Scope and proposal

    A short written proposal: approach, milestones, fixed price or retainer, and risks.

  3. Step 3

    Build with weekly demos

    Working software every week, decisions recorded as ADRs, evals and monitoring from day one.

  4. Step 4

    Handover

    Docs, runbooks and a walkthrough so your team owns it. Optional support after launch.

Ask my AI

Questions before a call? Ask the site.

The assistant answers from the content of this website only, and links to the page each answer comes from. It is a small working example of the grounded AI I build for clients.

How the assistant works →

What clients say

Testimonials

[Client quote about the outcome you delivered, ideally with a number.]
[Name, role, company]
[Client quote about how working with you felt: clarity, speed, communication.]
[Name, role, company]

Have an AI feature to design or ship?

Book a free 30-minute discovery call. You will leave with a clear next step, whether or not we work together.

Book a discovery call