Software for institutions
that cannot afford to be
wrong.
Placement readiness. Emergency care telemetry. Clinical examination. Autonomous finance. Four domains, one discipline — every number these systems put on a screen can be opened up and traced back to whatever produced it.
- 01
- 02
- 02
- 02
Every system we build
shows its working.
This was not a principle adopted up front. It is a pattern noticed afterwards — separate products, built at different times for different buyers, each independently refusing to claim what it could not evidence.
Every readiness score opens into the components behind it. Never a black box.
A student nobody has assessed is flagged unassessed — never quietly counted as safe.
Indicative requirement profiles based on publicly known hiring patterns, not official company cutoffs.
Argues, in writing, for everything it decides not to chase — and counts the contacts it avoided.
Provenance from any claim the interface makes back to the source record that produced it.
Telemetry only. No AI diagnosis — a deliberate limit, not a feature we have not reached yet.
AimPlacement
Placement readiness, measured
Every student gets a personalised prep path. Faculty, HODs, placement officers and the principal get one honest, shared view of how placement-ready the institution actually is — the same underlying numbers, rolled up to whoever is looking.
- 55,578
- 81
- 204
- 62
“6 companies. Here’s the exact gap.”
The same underlying numbers, rolled up to whoever is looking. A faculty member sees their classes, a HOD their department, and no more.
- AI prep assistant
- Interview simulator
Listed because they are real work in progress, and separated because a college choosing a pilot should be told exactly what it is getting today.
A college runs on eight
disconnected systems.
ERP here, attendance there, a placement portal, an LMS, library software, finance, HR, and the WhatsApp groups doing the work none of them do. The long project is to replace that sprawl one module at a time — starting with the one institutions feel most.
- Placements
- Examinations
- Attendance
- Academics
- Finance
- Library
- Research
- Administration
AimPlacement is module 01, and it is already in colleges. Everything greyed out above is named, not begun — this is the direction, not a roadmap with dates on it.
Two clusters,
one discipline.
AimPlacement is the one that ships today. These are the rest, at the stage they are actually at — named honestly, because a prototype described as a product is the first thing an investor checks.
Institutions
ECDEP
Live vitals from the ambulance, before the ambulance arrives
Streams a patient's vital signs and GPS from an ambulance to the receiving hospital's dashboard in real time, replacing voice-and-radio handoff with continuous data. Sensors to edge gateway to cloud to dashboard, on one data contract — so swapping simulated sensors for a real hospital monitor is additive, never a rebuild.
Telemetry only. No AI diagnosis, no clinical recommendation — a deliberate choice to stay outside medical-device classification rather than a feature we have not got to yet.
DigOSCE
Reveal the clinical case one stage at a time
A dual-view tool for OSCE-style clinical exams. Faculty drive from a tablet, releasing history, then vitals, then labs, then imaging; the projector shows students only what has been revealed. Built on the AimPlacement stack and sharing its identity layer.
The student view can never un-reveal something already shown, even through a dropped connection — the reconnection path is the design, not an afterthought.
HealthContext
A patient-continuity layer above hospital records
For multi-hospital groups, sitting above the incumbent HIS and EMR rather than replacing them. The flagship idea is "what changed" — the previous-visit-versus-this-visit comparison that nobody currently assembles.
An evidence chain is the first requirement, not the last: every claim the interface makes traces back to the source record that produced it.
Applied AI
Autonomous Finance OS
Not finance software — an AI finance officer
Phase one is built end to end: invoice intelligence, accounts receivable and expense intelligence. An invoice moves through extract, duplicate check, validate, PO match, decide, then persist with an audit record.
Every decision persists its own audit trail, and the agent asks for human approval wherever policy requires it rather than wherever it happens to be unsure.
RecoverOS
Revenue recovery that argues for what it decides not to chase
Every dunning system asks how fast it can retry. RecoverOS asks which of these is worth touching at all — then shows its working. Built for the Razorpay AI Buildathon.
Each decision stores its own evidence — the arithmetic, the probability basis, every policy rule checked. The console renders that record and never recomputes it. The metric nobody else puts on a dashboard: contacts avoided.
Four rules, and the
reason for each.
Claim only what is verifiable
If a number cannot be pointed at production, it does not go on a page. This site says 55,578 questions because that figure was checked on 13 September; it says nothing about student numbers or placement rates, because no outcome data exists yet to support them.
State the limit out loud
ECDEP carries no diagnosis and no clinical recommendation, on purpose. A stated boundary is worth more to a hospital than an implied capability, and it is the difference between a telemetry product and a regulated medical device.
Show the components
Any score the software displays can be opened up and traced to the parts that produced it. A number a user cannot interrogate is a number they are being asked to take on faith.
One thing in production beats six half-built
Sequenced, not parallel. That is partly discipline and partly arithmetic — there is one pair of hands, so the only way to have something real is to finish it before starting the next.
Jeffin Sajan
Everything on this page — the platform live in colleges, the ambulance telemetry rig, the finance agents, the clinical exam tool — was built by one person, pre-funding, alongside a final year of engineering.
That is not stated as a boast. It is the constraint that shapes the work: no capacity to run six projects at once means each one has to earn its turn, and the one in production stays in production.
If any of this is
worth a conversation.
Partnerships, pilots at institutional level, and investment. For an AimPlacement pilot specifically, that product has its own site and its own process.