B-CON AI · reasoning · Language and reasoning
B-CON Brain
Writes the note, drafts the plan, answers the question.
The language model behind B-CON AI. It turns a transcribed consultation into a structured clinical note, drafts an assessment and plan for the doctor to review, suggests investigations and medicines against the hospital's own catalogue, and answers questions from live hospital data in plain Nepali or English.
Capabilities
What it does.
- 01Structured SOAP notes from a Nepali or English conversation
- 02Disease diagnosis support and a 7-step differential diagnosis, always reviewed and signed by the doctor
- 03Lab test and medication recommendations from NIH, CDC and Nepal Standard Treatment Protocol knowledge, matched to the hospital catalogue
- 04Lab, imaging and medicine suggestions matched to what the hospital actually stocks
- 05Ask B-CON: answers about today's OPD, revenue, stock or beds from live data
- 06Schema-constrained output, so every suggestion lands in the right field
- 07A safety layer that blocks unsupported claims and flags red-flag symptoms
Specification
How it is built and served.
- Foundation
- Open-weights mixture-of-experts model, 26 billion parameters, about 4 billion active per token
- Knowledge
- Built over three years from NIH and CDC clinical references, eHMIS data from Nepal, Nepal Standard Treatment Protocols, national formularies and each hospital's catalogue
- Tuning
- Clinical prompting, structured differential diagnosis, a safety rulebook
- Output
- JSON schema enforced, so results are structured, not free text
- Context
- 32k tokens: the visit, the history, the catalogue
- Serving
- Self-hosted GPU, OpenAI-compatible API, scales to zero when idle
- Privacy
- Identifiers removed before inference; nothing is used to train anything
Design principle
The doctor decides. The model drafts.
Nothing B-CON Brain writes reaches a patient record without a clinician reading it and signing it. Suggestions are matched to the hospital's own catalogue so they are orderable, and drug interactions come from a curated database, never from the model's memory.
One consultation, start to finish
- 1
Listen
B-CON Ears transcribes the visit as it happens.
- 2
Structure
B-CON Brain separates complaint, history, findings and the doctor's decisions.
- 3
Draft
Assessment and plan appear in the note with suggested orders from the catalogue.
- 4
Check
The rulebook checks interactions, allergies, doses by weight and red flags.
- 5
Sign
The doctor edits, signs, and the prescription prints in Nepali and English.
Where it runs
B-CON Brain is inside
- B-CON EHRHospital information and management, on one patient record.
- B-CON EHMIS550+ health facilities. Every HMIS register. DHIS2, automatically.
- B-CON PHRYour health, with you.
- B-CON PharmacyRetail, wholesale and chain pharmacy on one fast counter.
- B-CON Blood BankNepal's live blood network, from donor to patient.
The other senses