Surendra Allam

Senior AI engineer · Bengaluru, India

AI features that hold up in production.

I build AI agents, knowledge assistants and voice agents into products that already have users. Then I measure them: accuracy, cost per answer, latency.

Book a 20-minute call See the work

Services

What I build

AI features in your product

LLM features added to an existing app or SaaS, or a stuck AI MVP finished and launched.

Claude · OpenAI · Amazon Bedrock

AI agents that do real work

Tool-calling agents connected to your CRM, Slack, helpdesk or database, for lead qualification, support triage and internal assistants.

Python · FastAPI · LangChain

Knowledge assistants

Answers grounded in your documents and data, with citations and accuracy checks.

Retrieval · citations · evaluations

Ask your data

Plain-English questions turned into SQL or graph queries over your database.

SQL · Neo4j

LLM cost and latency

Prompt caching, model routing and evaluations, so quality holds while the bill falls.

Caching · routing · evals

Voice agents

An inbound receptionist or outbound qualification, in English and Hindi or Hinglish.

Retell · Vapi · Bolna · LiveKit

Work

Selected work

Natural-language data assistant

A data assistant that answers in plain English

What it does
People ask questions in plain English. The assistant turns them into graph and database queries, runs them, and answers from the data that comes back.
Where
Three production products.
Result
Live in all three products
Stack
Amazon Bedrock · Neo4j

AI-first engineering

Moving a team to AI-first delivery

What changed
The engineering team I lead moved to AI-first development with Claude Code.
Where
A team shipping three AI products.
Result
Feature delivery went from weeks to days
Stack
Claude Code

Open source · benchmarked

Ask your data, with a model benchmark

What it does
A Claude agent answers business questions with read-only SQL, applies governed metric definitions, asks when a name is ambiguous, and shows the query behind every number.
Proof
50 test questions, 3 runs each, graded by re-running the SQL.
Result
Sonnet 5.5 and Opus 5.5: 100%. Haiku 5.5: 96.7% at $0.67 per 1,000 questions
Code
github.com/surendrallam/ask-your-data-agent

How I work

Fixed price, a demo early, a clean handover

  1. 20-minute callThen a one-page spec with acceptance criteria.
  2. Fixed priceSplit into two or three milestones.
  3. Working demoDelivered in the first milestone.
  4. Production buildWith tests, deployed.
  5. HandoverREADME, video walkthrough and 14 days of fixes. Optional monthly support after that.

About

Ten years of AI engineering, from devices to LLMs

I've spent over a decade building AI that runs in the real world, from on-device deep learning to LLM applications. My research on deep-learning object detection is published by IEEE.

Today I lead a team shipping three AI products for customers in the US and India, built AI-first with Claude Code.

I also write The Evolution of AI, a weekly newsletter that tells the story of AI so anyone can follow it.

I take on a small number of freelance projects at a time, so each one gets my full attention.

Contact

Tell me what you want to build

Send your use case and I'll reply with a scoped plan and a fixed price. My working hours overlap the US working day and UK evenings.