10decoders runs a working R&D practice — not a slide deck. Signal processing, medical imaging, agricultural sensing, computer vision and applied AI pipelines, taken from hypothesis to a system a client can put in front of real users.
Trusted by leading enterprises and healthcare teams
Manikandan has spent his career at the point where hardware, signal and model meet — the part of engineering most software companies quietly outsource. Drones taught him the discipline: a flight controller does not forgive a lazy filter, and a sensor reading that is 40 milliseconds late is a crash, not a bug report.
He carried that standard into everything since. A spirometer waveform, a retinal photograph, a soil moisture curve under a paddy field, a video frame from a factory floor — they are all the same problem wearing different clothes. Get the signal clean, understand the physics underneath it, then let the model do the narrow thing it is actually good at.
Outside the lab he writes white papers, and he spends a meaningful share of his year in engineering colleges talking to students who have never met anyone who does this work for a living. He treats that as part of the job, not a favour.
Each one started as a question someone thought was unanswerable at our price point. Each one is now running code.
Lung function measurement is a signal problem dressed up as a medical device. We work on the parts that decide whether a reading is trustworthy: flow-volume curve reconstruction, noise and artefact rejection, drift correction, and effort quality detection — so a clinician sees a result they can act on rather than a number that needs interpreting.
Screening is the bottleneck, not diagnosis. Our work targets retinal image quality assessment, lesion detection and grading — with an obsession about the cases the model should refuse to call. A screening system that knows when to escalate to a human is worth more than one with a better headline accuracy number.
Most enterprise AI fails on architecture, not intelligence. This track is about matching the model to the job: when a small fine-tuned model beats a frontier one, where retrieval belongs, how evaluation is designed before a line of code is written, and what the pipeline costs to run at volume on day 400 rather than day 4.
Rice is grown on intuition and irrigated on habit. We work on soil moisture estimation from low-cost sensing, calibration against real field conditions, and correlating moisture profiles with growth stage — the groundwork for telling a farmer when to irrigate instead of telling them what a dashboard says.
An open track, and deliberately so. Detection and tracking under conditions the papers never test — bad light, motion blur, cheap optics, aerial angles from drone work. Video analytics that runs at the edge instead of shipping every frame to a GPU somewhere. Some of this graduates into VuFindr. Some of it stays an experiment, which is the point of having a track for it.
Healthcare still moves on paper that has been printed, annotated by hand, faxed and scanned before any system sees it. The easy part is reading the characters. The hard part is reading the document: which box on a claim form that number belongs to, which table row survived the page break, which pen mark overrides the printed value.
So the track has two halves. Layout-aware extraction on genuinely degraded input — deskew, low-DPI faxes, stamps and handwriting over print, tables that split across pages. Then a semantic layer that maps what came off the page to the concepts a payer or provider system actually consumes: member, plan, provider identifier, service line, date of service, denial reason. Character accuracy is not the metric. Whether the field was right is.
The other half of being right is knowing when you are not. We spend as much time on the confidence model — which fields the system is allowed to fill on its own and which get routed to a person — as on extraction itself. This is the research line behind DocuFindr.
Long before any of this had a client attached to it, Mani was building and flying things that fall out of the sky when you get the maths wrong. No dashboards, no retries, no forgiving user — just a machine in the air holding you to your own assumptions.
That is where the habits came from. Trust the sensor only after you have characterised it. Budget for latency before you budget for features. Test in the ugly conditions, because the ugly conditions are the real ones.
Everything on this page inherits from that.
Work that stays inside a company is a hobby. Three ways ours gets out.
Still an active interest, not a memory. Flight systems, sensor fusion and aerial imaging feed our work on edge compute and vision under imperfect conditions — and they are how we teach it to students who want to see something fly.
We publish what we learn, including the parts that did not work. Written for engineers and technical buyers, not for a marketing funnel.
Mani speaks at engineering colleges across Tamil Nadu and beyond — showing students what applied R&D actually looks like on a Tuesday, and what it takes to get there.
R&D at 10decoders is a standing team, not a rotation of whoever is on the bench this quarter.
Honest answers about how R&D actually works at 10decoders — and what it means for your engagement.
The R&D/Innovation Lab invests roughly 10% of the engineering team’s time building reusable accelerators ahead of client need — migration templates, cloud-cost-optimization modules, and data-platform connectors — rather than research for its own sake disconnected from client delivery.
Because accelerators are built ahead of specific client need, new engagements often start with roughly 30% of foundational work already done for common problem patterns the lab has already solved — a direct, measurable effect on timeline for those overlapping use cases.
Five areas: AI/ML adoption frameworks, legacy modernization accelerators, a unified data platform with pre-built connectors, cloud migration templates, and microservices/API scaffolding — covering the practice areas most relevant to 10decoders’ core client base.
Both — R&D has produced client-facing products including CheiAI, DocuFindr, and VuFindr, alongside internal accelerators not sold as standalone products, indicating the R&D investment extends beyond internal efficiency into commercialized IP.
Research partnerships, sponsored R&D, proof-of-concept builds, or a campus session for your students. Tell us what you are trying to find out and we will tell you honestly whether we are the right team for it.
Manikandan leads 10decoders' R&D practice — spanning digital signal processing, medical imaging, agricultural sensing, and computer vision pipelines. He'll be straight with you about what's worth doing and what isn't.
Love what we're doing? Want to partner and sell our products or services?
Explore partner programs →We reply within one business day.
Explore our latest engineering insights, architectural blueprints, and practical playbooks.
Most QA programs test each AI agent's output in isolation. Almost none test what one agent tells the next, which is exactly where multi-agent pipelines break in production. Here is the handoff-testing gap and how to close it.
AI-led delivery pods have already made some GCCs 30% flatter, and could remove half of middle management by the end of 2026. Job descriptions, pay bands, and employment records rarely catch up. Here is where that gap turns into real legal and retention risk.
67% of enterprise RAG teams select a vector database based on team familiarity or vendor marketing rather than use-case benchmarking. 54% switch within 18 months. Here is the evaluation framework that prevents the switch.
10decoders