Software products & services

Web and mobile apps, internal tools, automation, and AI where it actually helps. Built for other teams, and for ourselves.

Start a project How we work
00 / What we build

Small tools. Real problems. No ceremony.

Most software gets built, shipped, and quietly abandoned. We work in the opposite direction: prove the thing is needed, build the smallest version that works, then keep it alive.

Our own products

Software we build and run in-house. Focused tools that solve one problem properly instead of forty problems badly.

Automation & AI

Systems that take the repetitive work off a team's hands, connected to the tools they already use, and monitored so they don't rot.

Client builds

A small number of projects each quarter for teams who need something built properly and shipped fast, by people who've already made the mistakes.

Web
ReactNext.jsVueSvelteAstroTypeScriptTailwind
Mobile
React NativeFlutterSwiftKotlin
Backend
.NETNodePythonFastAPIDjangoGraphQL
Data
PostgreSQLMongoDBRedisOpenSearchKafkaSupabase
Infrastructure
AWSAzureDockerKubernetesTerraformNginxGitHub ActionsVercelCloudflare
AI
OpenAIAnthropicPyTorchHugging FaceLangChainOllama
00 / Products

Our own software.

Four things we built and still run. Each one started as a problem somebody actually had, not a feature list.

ARGUS

AI support co-analyst

An AI support co-analyst that answers from your own documentation and live operational data, and returns nothing rather than guessing when it cannot ground the answer.

Retrieval
PDF ingestion, semantic chunking with overlap, all-MiniLM-L6-v2 embeddings at 384 dimensions
Index
OpenSearch k-NN vector index with HNSW cosine similarity
Hybrid context
Vector search combined with live keyword-triggered SQL against PostgreSQL
Grounding
Top-k tuned against a cosine-similarity relevance threshold, so weak matches suppress the answer instead of inventing one
Inference
Llama-4 through an OpenAI-compatible vLLM endpoint, streamed token by token over Server-Sent Events
Operations
Prompt assembly, multi-turn session management, retrieval-score logging, health endpoints
Stack
Python · FastAPI · OpenSearch · vLLM · PostgreSQL
Try it: ask about this page Keyword similarity over this page's text. The product uses MiniLM embeddings and OpenSearch k-NN.

CodeCrate

Secure code execution

Runs untrusted code without trusting it. Every submission executes inside its own disposable container, so a bad program can only break its own sandbox.

Execution
Multiple languages, each in an isolated Docker container per submission
API
REST endpoints for code submission, execution management and result retrieval
Pipeline
GitHub Actions automates testing, image builds and deployment to EC2
Stack
Python · Docker · GitHub Actions · AWS

Catalyst1122

HR system for Rescue1122

Built for the government rescue service. Admin and employee portals covering payroll, attendance, leave, tasks and announcements, with the repetitive parts handed to models.

Portals
Admin and employee, covering payroll, attendance, leave, tasks and announcements
Models
Résumé parsing, face-recognition attendance, and an HR chatbot
Outcome
Manual HR processing time cut by 40%
Role
Lead full-stack developer
Stack
.NET WebAPI · React · ML

Multi-cloud CI/CD

One pipeline, two clouds

A full-stack app deployed to Azure and AWS from the same pipeline, so neither cloud is the one that everything depends on.

Targets
Azure App Services and AWS S3, both from a single build
Pipeline
Build, test and deploy automated end to end in GitHub Actions
Releases
Containerised, triggered by GitHub webhooks
Stack
.NET WebAPI · Azure · AWS · GitHub Actions
Most projects don't fail. They stall, and a stalled build burns the same money as one that ships.
00 / How we work

Evidence before enthusiasm.

Four steps, in order, on every project we take. The order is the point. Most software fails because someone skipped the first one.

01

Test the assumption

Find the belief the whole thing rests on, and try to break it before anyone writes code.

02

Build the smallest version

The least software that can answer the question honestly. Scope is cut, not padded.

03

Ship it early

Into real hands, in weeks. Nothing is learned from software nobody has used yet.

04

Delete what doesn't earn it

Every feature keeps its place by being used. The rest comes out.

00 / Case studyExample

Nine months of roadmap. Nothing in production.

A logistics operator running support with a team of forty. Every ticket was read twice: once to route it, once to answer it. The AI roadmap had been approved three quarters running.

  1. Stalled

    Eleven features, no first one

    Routing, summarising, drafting and a chatbot were all scheduled for the same quarter. Nobody could start one without a decision about the other three, so nobody started.

  2. Moved

    We built the router, for one queue

    The whole roadmap rested on one untested claim: that routing was the expensive half. We checked it against six weeks of their own tickets before writing any product code.

  3. Shipped

    One queue in 19 days, four more by month two

    Classification against the tags they already used. No fine-tuning, no chatbot. It has run since without a person watching it.

First version live
19 days
Roadmap features cut
8 of 11
Running unattended since
March 2026
00 / Client work

Bring us something stuck.

We take on a handful of client projects at a time so each one gets built properly. If you have an idea that needs proving, a prototype that needs to become real, or a system that keeps stalling, that's our work.

Capacity / Q3 2026 00 of 4 slots open
Tell us about it
  • 00Web and mobile applications, end to end
  • 00MVPs built and shipped in weeks, not quarters
  • 00Internal tools and workflow automation
  • 00Idea validation and technical feasibility
  • 00AI and LLM features, where they earn their place
  • 00Rescue work on builds that have stalled
Salman Javed
00 / Founder

Salman Javed

CEO & Founder

Salman founded Deadloc AI to build software that holds up after launch, the part most projects skip. He sets product direction and stays close to the engineering, so the decisions about what ships are made by someone who has to live with them.

His most recent build is ARGUS, an AI support co-analyst whose most useful property is knowing when not to answer: it grounds responses in a customer's own documentation and live operational data, and returns nothing rather than guessing.

Contact

One email.

That's the whole process.

Tell us what you're building and what's in the way. We reply to everything, usually the same day.

hello@deadlocai.com

Already working with us? support@deadlocai.comLinkedIn

Prefer a form?