Codebase review & roadmap
We analyse your repository, tests and delivery path, then hand back a prioritised plan: what to fix, what to build, what to leave.
AI research & engineering
Practical intelligence.
Rigorous engineering.
We build and investigate AI systems that understand context, use tools and act in digital and physical environments.
01 · Build
Engineering for new companies
Ship and harden your product with Adam, our agent harness, plus senior engineering judgement.
02 · Back
Ghost pilots & funding
Bring a physical task, sponsor an evaluation or invest in efficient physical intelligence.
03 · Collaborate
Research partnerships
Joint questions, reproducible experiments and collaborative funding proposals.
Our research focuses on efficient physical intelligence and dependable software agents, combining working prototypes with controlled experiments and measurable evaluation.
Research · 01
Ghost
Efficient intelligence for physical tasks
Ghost explores how visual understanding and compact recurrent controllers can work together to support physical tasks.
Our early work combines video-based task understanding, vision-language models and simulation workflows. We have prototyped bio-inspired recurrent controllers and compared them with a GRU baseline.
The next research stage will investigate robustness under changing observations, recovery from failures, and the relationship between task performance, latency and energy use.
Status: research prototype and experimental evaluation
Also: How machines learn to touch, a short film about Ghost ↗
How much intelligence can we deliver within a small computational budget?
Our central question
Work with Ghost
Help us take Ghost from prototype to proof.
We are looking for partners who bring real tasks, data or funding, and want honest numbers back.
- Sponsored evaluation Bring a manipulation or handling task. We model it in simulation and report task success, recovery, latency and energy against a baseline.
- Pilot partnership Robotics, automation and hardware teams: test compact controllers on your platform and shape what we build next.
- Investment & grants Investors and funding bodies interested in efficient physical intelligence: we share a technical brief and a research plan on request.
Research · 02
Adam
Engineering agents with controlled execution
The agent does the legwork. A person makes the call.
Adam
Branch ready: invoice parser, 14 tests passing. Waiting for your review.
You
Reviewing now. Hold the deploy until I approve.
Adam is our custom engineering harness for repository analysis, implementation, automated validation and controlled software delivery.
It combines language-model orchestration and context management with testing, Git workflows and deployment handoffs. Autonomous development is separated from production credentials.
Our research direction explores how agents maintain useful context as tasks evolve, verify their work and operate within explicit permissions and human review boundaries.
Status: internal engineering tooling with ongoing development
Agent sandbox · no production credentials
1
AnalyseRead the repository, map the change, gather context
2
ImplementSmall, reviewable commits on a branch
3
ValidateTests, type checks and builds must pass
Human review
4
Hand offMerge and deploy through your own pipeline and keys
Adam for new companies
Engineering capacity, without the hiring lag.
Early teams need software that ships and holds up. We pair Adam's throughput with senior engineering judgement, so you get working features, tests and a clean handoff, in your repository, under your control.
Prototype to product
Turn a demo into something customers can rely on: features built with tests, CI, error handling and deployment handoff.
Agent-ready engineering
Set your team up to use coding agents safely: test harnesses, review gates, permission boundaries and credential separation.
Ongoing delivery
A standing engineering partner for your roadmap, with regular reviews and every change traceable to a pull request.
What you can count on
- Work lands as pull requests in your repository
- Tests and checks with every change
- Agents never hold your production credentials
- A human reviews before anything merges
- Documented handoff, so your team owns the result
A good fit if you are
- A technical founder who needs more hands
- A non-technical founder who needs a build partner
- Building with embedded, automotive or robotics software
- Adding AI agents or LLM features to a real product
How we work
Build. Measure. Understand.
Build
We turn research questions into working prototypes, not slideware.
Measure
We compare approaches against relevant baselines in controlled experiments.
Understand
We examine where systems fail and document the conditions under which they work.
Our approach brings together software engineering, systems modelling, automated testing and experimental evaluation. We distinguish demonstrated results from hypotheses.
Research collaboration
Let's design the experiment together.
We welcome conversations with university researchers and industry partners working on:
- Efficient AI
- Robotics
- Recurrent architectures
- Agent evaluation
- Dependable autonomous systems
- Joint research questionsScoped so both sides learn something publishable or useful.
- Reproducible experimentsShared baselines, documented conditions, honest failure cases.
- Practical demonstratorsPrototypes that show a result working, not just a chart.
- Collaborative funding proposalsGrant consortia and industry-academic programmes where an engineering partner strengthens the bid.
About RNDM Labs
A systems view, hands on the code.
RNDM Labs is an independent AI research and engineering lab.
Founded by Vijay Lakshminarayanan, our work draws on more than 15 years of engineering experience across embedded systems, automotive technology and AI, including systems engineering and software quality at Bentley Motors.
We combine hands-on implementation with a systems perspective, connecting model behaviour to the software, infrastructure and controls needed to make it useful.
- 15+
- years in engineering
- Embedded · Automotive · AI
- domains
- Ghost · Adam
- research programmes
Start a conversation
Tell us what you're working on.
Pick a topic and add a few lines. This opens a ready-to-send email in your own mail app; nothing is stored on this site.