Custom software & applied AI

Software that fits you. Not the other way around.

Every business works differently. Off-the-shelf SaaS asks you to bend to it. We build software around your exact requirements, at a cost that makes custom realistic for everyone, not just enterprises.

Bangladesh, working worldwide
Research-led, deployed alongside you

Our approach

We are not SaaS.

SaaS is built for the average customer, and nobody is the average customer. We start from what you need and build exactly that. Nothing you have to work around, nothing you pay for and never use.

Off-the-shelf SaaS One size

  • You adapt your workflow to the product
  • Features built for everyone else's needs
  • Per-seat fees that grow with you, forever
  • Your request sits on someone else's roadmap

Ediphas Labs Made for you

  • Software shaped around how you already work
  • Only the features you actually need
  • Low-cost custom builds, priced for real budgets
  • Your requirements set the roadmap

Services

What we build

From a simple internal tool to a full AI system, every project starts with your requirements and ends with software that fits them.

AI & LLM apps and agents

Assistants, agents and AI workflows built on your own data and processes, designed to be reliable, not just impressive in a demo.

RAGAgentsChatbotsAutomation

Full-stack web development

Web apps, dashboards, portals and APIs. Business systems like ERP, registration and operations platforms, built to your workflow.

Web appsDashboardsAPIsERP

Applied AI research & R&D

Prototypes, feasibility studies, model evaluation and benchmarks. Find out what AI can really do for your problem before you commit.

PrototypesEvaluationBenchmarks

Data & ML engineering

Datasets, pipelines, model training and deployment. Turning messy real-world data into something your systems can use.

PipelinesDatasetsTrainingMLOps

How we work

Research-led. Deployed alongside you.

We bring a researcher's rigour to every build, and we don't disappear after hand-off. We work inside your context until the software truly fits.

01

Understand

We map your requirements, workflow and constraints before writing a line of code.

02

Prototype

A working version early, so you can react to something real instead of a document.

03

Build & deploy

We build it, deploy it in your environment and adjust it with your team.

04

Evolve

As your needs change, the software changes with them.

Selected work

Things we've built

A selection of systems, tools and platforms from our engineering and research work.

Event platform

IUT Fest

Registration and operations platform for a six-event university festival.

LaravelMySQLRedis

AI agent

Pi Daytona Chat

Coding agent where every conversation gets its own isolated sandbox, and every tool call is logged.

TypeScriptConvexDaytona

AI education

ISHKAPON AI

Local-first desktop tutor that computes every number in a sandboxed Python runtime instead of trusting the model's memory.

ElectronReactPyodide

Developer tool

Reko

Extracts every function from a codebase in 15 languages into one representation, with byte-exact reconstruction and semantic search.

RustONNXsqlite-vec

Generative AI

AIManim

Turns a plain-language maths prompt into a narrated, animated explainer video.

PythonManimLLMs

Open data

Bangladesh Legal Acts

Every Bangladeshi act from 1799 to 2025, structured and bilingual: 1,484 acts and 35,633 sections.

PythonData engineering

Research

Research that ships

Our work on LLM evaluation, safety and low-resource language AI is published at leading venues. The same rigour goes into everything we build for clients.

Best Paper Award · AACL-IJCNLP 2026

LLM-Ideoplasticity

Measuring how the political behaviour of large language models shifts with context.

LREC 2026

Figurative understanding in Bengali idioms: a 10,361-idiom benchmark tested on 30 LLMs.

ICML 2026 · MusIML

AREG: an adversarial game for evaluating persuasion and resistance in LLMs.

EMNLP 2026 · REALM & MRL

Information containment in multi-turn LLM dialogue, and what multilingual benchmarks really measure.

12

Papers and preprints on LLMs and NLP

5,900+

Downloads of our open legal dataset on Kaggle

15

Programming languages parsed by Reko

Your requirements. Your software.

Custom solutions that fit you, without adjusting to everyone else.