Great project - the multi-agent architecture with specialized agents and human-in-the-loop validation is solid work.
One area that comes up a lot in patient interactions: supplement-drug interactions. When a patient mentions they're taking ashwagandha or fish oil alongside prescribed medications, the assistant currently has no structured data source to check for interactions. LLMs tend to either miss these or give vague "consult your doctor" responses without specifics.
We've created a free API called TruthStack that provides structured supplement-drug interaction data:
- 1,008 directed interactions with severity and evidence scoring
- FDA adverse event signals (FAERS + CAERS)
- 95 compounds with fuzzy alias matching (catches "mag glycinate" = magnesium, "KSM-66" = ashwagandha, etc.)
It could plug in as a tool for one of your existing agents, or as a dedicated safety-check step in the workflow. Example:
pythonimport requests
result = requests.post(
"https://api.truthstack.co/api/interactions/check",
headers={"X-API-Key": "your-key"},
json={
"supplements": ["st johns wort", "magnesium"],
"medications": ["sertraline"]
}
)
Returns: CRITICAL risk — St. John's Wort + sertraline = serotonin syndrome risk
There's also an MCP server if you want agent-native tool discovery: https://github.com/TruthStack1/truthstack-mcp
Just a suggestion and I'm happy to answer questions if it seems useful.
Great project - the multi-agent architecture with specialized agents and human-in-the-loop validation is solid work.
One area that comes up a lot in patient interactions: supplement-drug interactions. When a patient mentions they're taking ashwagandha or fish oil alongside prescribed medications, the assistant currently has no structured data source to check for interactions. LLMs tend to either miss these or give vague "consult your doctor" responses without specifics.
We've created a free API called TruthStack that provides structured supplement-drug interaction data:
It could plug in as a tool for one of your existing agents, or as a dedicated safety-check step in the workflow. Example:
pythonimport requests
result = requests.post(
"https://api.truthstack.co/api/interactions/check",
headers={"X-API-Key": "your-key"},
json={
"supplements": ["st johns wort", "magnesium"],
"medications": ["sertraline"]
}
)
Returns: CRITICAL risk — St. John's Wort + sertraline = serotonin syndrome risk
There's also an MCP server if you want agent-native tool discovery: https://github.com/TruthStack1/truthstack-mcp
Just a suggestion and I'm happy to answer questions if it seems useful.