AI/ML & Data
Quantising vision models for sub-5 W edge inference
How much accuracy a detection model actually loses on the way from a workstation GPU to an embedded accelerator, measured across four quantisation strategies and three target boards.
Ten disciplines, one engineering organisation — each deep enough to stand alone, most valuable combined.
All capabilitiesIndustries
Twenty sectors, one engineering practice
Where failure has real consequences — and a multidisciplinary team earns its keep.
All industriesTraining & Education
Capability that stays with your team
Hands-on programmes taught by the engineers who do the work.
Training overviewProgrammes
Professional Training Structured programmes for teams and institutions Technical Courses Hands-on engineering courses, taught in the lab Workshops & Bootcamps Intensive short-format, hands-on sessionsSubject Tracks
Cybersecurity & Forensics VAPT, incident response and digital evidence AI/ML & Data Applied ML, computer vision and data engineering Software Engineering Clean code, APIs, testing and delivery Embedded & Electronics Firmware, IoT, PCB design and bring-up GIS & Geospatial Remote sensing, spatial analysis and WebGIS
Programmes for your team
Curricula built around a specific team, delivered on-site, at our lab or online.
Plan a programmeDwarka Mor, New Delhi
[email protected]
Home/Research
We publish what we learn. Benchmarks, methods and failure modes from live programmes — written for engineers, not for procurement.
Published Work
Each paper comes out of work we actually shipped. Where a result is inconvenient for us, it is in there too — a benchmark that only reports wins is marketing.
AI/ML & Data
How much accuracy a detection model actually loses on the way from a workstation GPU to an embedded accelerator, measured across four quantisation strategies and three target boards.
IoT & Embedded
A reference chain of trust for parts without a dedicated secure element, covering key storage, signature verification timing and the failure modes an attacker actually reaches for.
GIS & Spatial
A processing approach for monsoon-affected geographies where optical archives are unusable for months at a time, with accuracy measured against ground survey.
Cyber & Forensics
A practical method for teams shipping devices into OT environments: what to model, what to skip, and how to keep the model alive after the architecture freezes.
Simulation
Keeping a plant or network twin honest when sensors are few, noisy and occasionally offline — with calibration error measured against held-out field data.
Hardware
Grounding, return paths and stack-up choices that decide whether a board passes pre-compliance the first time, with measured before-and-after scans.
Papers are shared on request so we can answer questions alongside them. Tell us which one you need and what you are building.
If one of these touches what you are building, the fastest next step is a conversation with the engineer who wrote it.