Indic Labs — Studio

The Work.

Production agent systems and GenAI content pipelines. Everything below runs for a real client or in our own stack. No mockups in the metrics.

OPD Orchestrator — intake to bill, one graph.

opd-orchestrator.flow — production · v3.8 ACTIVE
TRIGGER
Patient Intake
webhook · kiosk + desk
uhid.create | abha.link
2.1k/day34ms
EHR Context
fetch: history, allergies
FHIR R4 · ABDM gateway
cache 91%118ms
LLM
Triage Agent
claude-sonnet · tools: 6
urgency + specialty route
eval 98.2%1.4s
Queue Router
slot: consult desk 1–9
wait-time prediction
p95 4m22ms
Bill Drafter
GST + TPA lookup
human approve gate
0 err/30d340ms
DONE
Visit Record
postgres + audit log
TRIGGER
Voice Intake
twilio · 4 languages
deepgram STT stream
610/dayreal-time
LLM
Symptom Parser
extract → structured
hand-off to triage
eval 97.1%900ms
CATCH
!Escalate Human
low-confidence → senior desk · slack alert
1.8% of runs
14:02:11 run #88,412 — intake→triage→queue · desk 4 · total 1.9s
14:02:09 triage_agent tool_call: ehr.fetch_history(uhid=SS-2041-…) 112ms
14:02:08 run #88,411 — voice intake (hi-IN) → symptom parse conf 0.94
14:01:56 run #88,407 — confidence 0.61 < 0.7 → escalated to senior desk
14:01:42 bill draft approved by operator@reception-2 · ₊ audit logged
2,700+runs / day
OPD throughput
98.2%triage eval score
1.8%human escalation
Read the Case Study →

Reel Factory — brief in, 30 reels out.

reel-factory.flow — production · v5.2 ACTIVE
CRON
Weekly Brief
mon 06:00 IST
notion: content_db
52 wks
LLM
Hook Writer
brand-voice RAG
30 scripts · hook+cta
top-hook loop40s
Magnific Render
img gen + upscale
brand LUT lock
30×/run4m
Cut + Caption
ffmpeg pipeline
karaoke captions · SFX
9:16 + 1:12m
SHIP
Scheduler
IG · YT · LI
best-slot model
JUDGE
Brand QA Gate
LLM judge: tone, claim,
visual compliance · blocks fails
rejects 7%6s
06:14:33 batch W24 — 30/30 rendered · 2 rejected by QA → re-gen · 28 scheduled
06:11:02 hook_writer: top performer last wk "nobody tells you this about cutting" → seeded 4 variants
06:08:47 reel_19 claim check: "2x faster fat loss" → flagged, rewritten compliant
06:02:10 analytics ingest: wk23 — avg watch 71% · saves +18% w/w
30reels / week
1human operator
3.1×engagement lift
7%QA rejection rate
Read the Case Study →

Every graph we run.

GenAI pieces, brand-locked.

Real renders land here as clients approve public use. Until then — structure over screenshots.

Sites that sell, dashboards that decide.

Want yours on this page?

Tell us the workflow. We'll show the agent. Tell us the brand. We'll show the engine.

Start a Conversation