# Dravin AI

> Dravin AI is an AI software company based in India that builds AI systems and the software around them: AI agents and agentic systems, voice agents, retrieval (RAG) and document AI, fine-tuned, pretrained and self-hosted models, inference optimisation, classical ML, generative media, and the web, mobile and desktop apps they ship in. We also audit and clean up AI-built apps that break in production. How we build: coding agents draft in parallel in isolated environments, CI tests and previews every pull request, one of our engineers approves each change, and every merge deploys automatically. Two years serving customers, 30+ production AI systems shipped, no public prices (scope and fee are agreed on a call).

Some of the companies we’ve worked with: Cognizant, Hewlett Packard Enterprise, Bhindi AI (now Mogra), HelloBible, Graspr AI, Altiv, Blinxed.
Start a project: book a call or email info@dravin.ai. The clean full text of this site is at https://dravin.ai/llms-full.txt.

## Pages

- [Home](https://dravin.ai/): what Dravin AI builds, how a change ships, the stack and the latest posts
- [Services](https://dravin.ai/services/): every capability in depth (agents, RAG, voice, training, self-hosting, inference, classical ML, generative media, apps, cloud, code audits) and the six ways to work with us
- [How we work](https://dravin.ai/how-we-work/): how an engagement starts, parallel coding agents with an engineer approving each change, automated checks, testing, previews, CI/CD, the intake board and handover
- [Work](https://dravin.ai/work/): some of the companies we have worked with and what we have built, grouped by capability
- [Company](https://dravin.ai/company/): who we are, the rules we hold every project to, and answers to the questions people ask before they hire us
- [Request a code review](https://dravin.ai/review/): a written review of one repository within 48 hours, with findings ranked by what to fix first
- [A full code review report](https://dravin.ai/review/sample/): a whole written review of a made-up app: the summary, findings by severity, the fix for the worst one, what to leave alone and what to fix first
- [Contact](https://dravin.ai/contact/): book a call or write to us

## What we build

- [Agents and automation](https://dravin.ai/services/#agents): Agents that take real work off a team: they plan, call your tools, remember what they did, run on a schedule, hand work to each other and wait for a person before anything risky. From one copilot to a platform driving 100+ agents from one chat window.
- [Content and generative media](https://dravin.ai/services/#media): Content engines that write, design, film and publish in your voice and in any language: SEO and AI-search content, social, newsletters, ads, product imagery, video and dubbing, with a person approving what goes out.
- [Voice agents](https://dravin.ai/services/#voice): Voice agents for support lines, bookings, sales calls and pre-screens: speech-to-speech and streaming pipelines that start speaking before the model has finished and stop when the caller cuts in.
- [Your own AI](https://dravin.ai/services/#training): Your own AI, trained on your data: continued pretraining, fine-tuning, RL post-training and distillation, from small models for a phone to open mixture-of-experts models with hundreds of billions of parameters, and 1B to 10B-class models pretrained from scratch.
- [Inference and on-device](https://dravin.ai/services/#self-hosted): Models served where your data lives: on the phone, the laptop or the browser, on one private GPU, or across a cluster, with routing, caching, batching, quantisation and speculative decoding that cut cost and latency where quality holds.
- [Beyond LLMs](https://dravin.ai/services/#beyond-llms): We use an LLM only where language is the job. When the answer is a label, a number, a forecast or a finding in an image, a small model trained on your data is faster, auditable, more stable and lighter to run.
- [Apps and business software](https://dravin.ai/services/#products): Full products and the software a business runs on: web and SaaS apps, iOS and Android apps, Mac, Windows and Linux apps, and CRMs, ERPs, inventory systems, self-hostable project trackers, learning platforms, portals, dashboards and internal tools, with AI built in where it helps.
- [Fixing AI-built apps](https://dravin.ai/services/#code-quality): We find what will break in production and fix it, including apps built with Lovable, Bolt, Cursor, Replit, v0 or Claude Code that now have real users.
- [Retrieval, memory and document AI](https://dravin.ai/services/#rag): Assistants that know your documents and remember past work, and pipelines that read invoices, contracts, specifications, claims and clinical notes into clean, checked data.
- [Evaluation, search and safety](https://dravin.ai/services/#evaluation): Scored harnesses that show whether a change made the model better, guard models and red-teaming before release, and search that answers with citations.
- [Cloud, DevOps and architecture](https://dravin.ai/services/#cloud): System architecture and deployment on AWS, GCP, Azure or your own data centre: containers, Kubernetes, infrastructure as code, GPU clusters and serving, CI/CD and observability.

## Ways to work with us

- [Project](https://dravin.ai/services/#project): A defined build with an end date.
- [Embedded engineering](https://dravin.ai/services/#embedded): We join your engineering team and work to your standards: your repository, CI, linters and code owners.
- [Ongoing engineering](https://dravin.ai/services/#ongoing): Steady engineering on a live product, with bugs and features dropped on one intake board.
- [Code audit](https://dravin.ai/services/#audit): Where the 48-hour report reads one repository, an audit covers every repository in scope and ends with a written plan for what to fix first.
- [Vibe-code cleanup](https://dravin.ai/services/#cleanup): An AI-built app made safe, tested and maintainable, usually without a rewrite.
- [Consulting](https://dravin.ai/services/#consulting): A second opinion on architecture, model choice and delivery for AI systems.

## Blog

- [Blog](https://dravin.ai/blog/): long reads on AI engineering and the AI services market, with the data behind them
- [More users per GPU: a playbook for self-hosted LLM serving](https://dravin.ai/blog/raising-llm-serving-throughput/): What to measure first, then the levers that raise LLM serving throughput on your own GPUs, in the order we pull them and what each one trades away.
- [Typed decision models: what early tests show](https://dravin.ai/blog/jev-and-typed-decision-models/): Models that answer typed questions with a probability in one pass: what they are, what the first independent tests found, and how to test one.
- [The 90 firms that fix AI-generated code](https://dravin.ai/blog/who-cleans-up-ai-generated-code/): What the 90 firms say breaks in AI-built apps, how they package the work, what none of them sells, and a checklist to use on any of them, including us.
- [What 843 AI agencies actually sell](https://dravin.ai/blog/what-843-ai-agencies-actually-sell/): We mapped 843 AI agencies across India, the US and Europe: what they sell, where they are, which offers are crowded, which are thin, and what to check.
- [RSS feed](https://dravin.ai/rss.xml): the newest posts in full, newest first
- [JSON Feed](https://dravin.ai/feed.json): the same posts as JSON Feed 1.1

## Optional

- [Security](https://dravin.ai/security/): what we do with access to your code
- [Privacy](https://dravin.ai/privacy/): what we collect, for how long, and your rights under the DPDP Act
- [Terms](https://dravin.ai/terms/): the terms that apply to client work and this website
- [Full text](https://dravin.ai/llms-full.txt): the main pages, the FAQ, the policies and every post as clean Markdown
- [Sitemap](https://dravin.ai/sitemap-index.xml): every indexable URL