Clinical AI that answers from the record, and shows you where.
SMRITI: State-and-Memory Reading Inference with Traceable Information
Ask a question about a patient in plain language. SMRITI reads the notes that matter, cites every answer, and tells you when the record doesn't say. Built for hospitals in India and the United States.
Ask the record
Fictional patient, for illustration
- 12 Feb 2022Pathology. Adenocarcinoma of the sigmoid colon.
- 03 Mar 2022Staging. TNM stage T3 N1 M0.
- 10 Mar 2022Biomarkers. KRAS wild type. Microsatellite stable.
- 21 Mar 2022Treatment. Started FOLFOX as first-line therapy.
- 19 Sep 2022Imaging. Partial response on first-line therapy.
- 06 Apr 2023Treatment. Started FOLFIRI with bevacizumab as second-line therapy.
- 11 Oct 2023Imaging. Stable disease on second-line therapy.
- 15 Feb 2024Treatment. Started cetuximab as third-line therapy.(read in step 2)
- 22 Aug 2024Imaging. Progression on third-line therapy.(read in step 1)
- 30 Aug 2024Lab. CEA rising compared with the previous result.
Answer
Third-line cetuximab
Found the scan that showed progression, then the treatment line it refers to.
Pick a question and watch which records SMRITI reads before it answers.
Same question. Two very different answers.
Most AI models answer from what they absorbed during training and re-read everything you give them. In healthcare, an answer is only useful if you can check it.
- Where knowledge lives
- Typical AI: Frozen inside the model at training time.
- SMRITI: In an editable memory, next to its source.
- Each new question
- Typical AI: Re-reads the whole history, so long charts cost more and get blurry.
- SMRITI: Looks up only the records the question needs.
- When the answer isn't there
- Typical AI: Often produces a fluent guess.
- SMRITI: Says it is not documented.
- Showing its work
- Typical AI: No reliable link back to a document.
- SMRITI: Every answer lists the record lines it used.
- New guideline or code update
- Typical AI: Waits for the next model release.
- SMRITI: Takes effect as soon as it is added.
- Where it runs
- Typical AI: Usually a third-party cloud.
- SMRITI: Inside your hospital or private cloud.
One model. Two healthcare systems.
Regulation, record standards and code sets differ between India and the US. SMRITI is set up for each, so answers and codes match the rules your hospital works under.
For hospitals, cancer centres and health programmes working within India's digital health framework.
- Regulation
- Designed for the DPDP Act. Patient data stays in your hospital or an Indian data centre.
- Health records
- Reads records from your HIS, LIS and PACS, including ABDM-linked records shared with patient consent.
- Code systems
- WHO ICD-10 for diagnoses, and SNOMED CT through India's national licence.
- Workflows
- Discharge summaries, insurance and TPA claim documentation, tumour board preparation, NABH documentation.
Same note, local code set
“Invasive carcinoma of the right breast. Quadrant not specified.”
WHO ICD-10
C50.9
Malignant neoplasm of breast, unspecified
WHO ICD-10, used in India, does not record which side.
Six letters, each one a design decision.
SMRITI is built around memory the way a careful clinician works: read the chart, find the relevant notes, connect them, and be clear about what isn't written down.
State
A compact working memory that holds what matters for the question in front of it, so the model never has to re-read the whole chart.
From a new report to a cited answer.
- 1
Write
Each note, report or code table is read once as it arrives, and stored with its source and date.
- 2
Remember
Everything is kept in a memory you control. Edit or remove a record and the next answer reflects it.
- 3
Read
For each question, SMRITI fetches the records it needs, following one to the next when an answer spans several.
- 4
Answer
It answers with the records it used, or says plainly that the record doesn't contain the answer.
Ten notes or ten years of notes. The effort per question stays the same.
Because SMRITI reads only what a question needs, a long and complicated history doesn't make each answer slower or more expensive. And when a guideline or code table changes, you update it once.
Update knowledge without retraining
Where SMRITI helps first.
Ask anything about a patient's history in plain language. SMRITI answers from the chart and shows the notes it used, across years of visits.
Increased to 1000 mg twice daily on 4 Jun 2025, after the HbA1c result was above the patient's target.
Not documented in this patient's record.
No diagnosis, lab result or note mentions kidney disease.
What SMRITI will and won't do.
Trust in clinical AI comes from clear limits. These are built into how SMRITI works, not added as settings.
It will
- Cite the record lines behind every answer
- Say “not documented” instead of guessing
- Keep patient data inside your environment
- Pick up new guidelines and code updates without retraining
- Keep a log of what it read for every answer
It won't
- Replace a clinician's judgement or sign-off
- Recall medical codes from memory
- Send patient records to an outside service
- Hide where an answer came from
Runs where your patient data already lives.
On your infrastructure
Runs on-premise or in your private cloud. Patient records stay where they are.
HIPAA, DPDP and ABDM ready
Designed for the privacy rules of both markets, alongside the rest of the Healytix platform.
Audit-ready by default
Every answer is stored with the questions asked and the records read, ready for review.
Access follows your policies
SMRITI only reads what the person asking is allowed to see.
What people ask us about SMRITI.
Something else on your mind? Write to contact@healytix.com.
Is SMRITI a chatbot?
No. It is a language model designed to answer from a specific record. You can ask it questions in plain language, but its answers are tied to the documents it reads, not to general knowledge.
Does it replace doctors or medical coders?
No. SMRITI prepares answers and suggestions with their evidence so that people can review them faster. Decisions and sign-off stay with your clinicians and coders.
Where does our patient data go?
Nowhere outside your environment. SMRITI is deployed inside your hospital or private cloud and reads records in place.
What happens when the record doesn't have the answer?
SMRITI says the information is not documented. It is designed to prefer an honest gap over a confident guess.
How do we keep it up to date?
Add the new guideline, protocol or code table to its memory. There is no retraining cycle to wait for.
Is SMRITI available in India and the US?
Yes. SMRITI is designed for both: WHO ICD-10, SNOMED CT, ABDM and the DPDP Act in India, and ICD-10-CM, HL7 FHIR and HIPAA in the United States. Beta partners in both countries are welcome.
Who can join the beta?
Hospitals, cancer centres, clinicians, researchers and partners who want to pilot SMRITI on real workflows and help shape it. Write to us and we will set up a conversation.
What happens on 5 November 2026?
Live demos open. Request a slot and we will walk you through SMRITI on example records, and discuss a pilot on yours.
Let's build it together.
SMRITIBeta is now open for early partners.
- Hospitals and cancer centres: pilot it on your workflows
- Clinicians and researchers: shape it with us
- Partners and investors: build it with us
SMRITI is a research preview. It supports clinician review and is not intended for diagnosis or treatment decisions. Patient examples on this page are fictional.