The element of examination
Elenchos — ἐλεγχος — is the Greek word for the Socratic cross-examination: the method by which Socrates did not tell a man what he knew, but questioned him until what he claimed to know either held or fell apart.
That is the entire product thesis in one word. Elenchium does not test recall. It asks, and asks again, and follows the answer down until comprehension either holds or reveals itself as a gap. A trainee who has memorised the right words does not survive an elenchus. A trainee who understands does.
The problem Elenchium addresses
Enterprise training has a cost curve that software has mostly failed to fix, because most software attacked the wrong layer. Human-delivered training scales linearly with headcount, and the labour cost of that headcount rises every year, while machine-generated instruction keeps getting cheaper. Almost nothing in the market is positioned on the correct side of that crossing.
Corporate training is three expensive humans in a trench coat — a content team that writes the material, a trainer who delivers it, and an examiner who certifies it. Incumbent tooling automates none of these roles, it only administers them: an LMS is a shell that tracks who watched what, and a content library licenses generic material that was never about your organisation.
Worse, what comes out the far end is a certificate, not evidence. A completion record cannot show whether a trainee understood the material or clicked through it — and when the decision downstream is employment-relevant, a certificate is precisely the wrong artefact. Language compounds the gap: the cohorts that most need structured onboarding are often the ones an English-only product cannot serve, and Tamil–English code-mixed delivery is the difference between a programme reaching a cohort and not.
Who Elenchium serves
- Enterprise and SME technical onboarding teams
- L&D and HR leaders who own role certification
- Engineering and professional colleges running placement-readiness programmes
- New hires and trainees themselves
- Managers who must make an employment-relevant decision and need an evidence packet to defend it
How Elenchium works
COURSE INTAKE
An administrator describes a course in five fields. From there, thirteen specialised agents take over: they plan the curriculum, write the lessons, build the decks, and generate practice sets and a unique exam per candidate, so assessment cannot be shared, memorised, or passed around a cohort. Every artefact passes a QA gate before it publishes.
LIVE VOICE CLASSROOM
A voice AI trainer teaches the class live, every day — not a recorded lecture, but a live classroom on a self-hosted voice stack, teaching in the language the cohort actually speaks, including Tamil–English code-mix.
EXAMINATION GATES PROGRESSION
Assessment is not a formality at the end; it is a gate. Trainees must clear a 70% threshold to progress, with two attempts and a 24-hour remediation window between them. Falling short twice escalates rather than silently passing a trainee along.
THE HUMAN DECIDES
Elenchium generates, teaches, examines and documents, assembling the full evidentiary record of what was asked, how it was answered, and where a trainee struggled or recovered — then hands that packet to a person. The system never decides whether someone is hired, confirmed, promoted, or held back.
What makes Elenchium different
The moat is not the agent pipeline — pipelines are copyable. The moat is the evidence packet: the record incumbent LMS platforms and content libraries cannot produce, because they never observed the examination in the first place.
| Category | Incumbent approach | Elenchium |
|---|---|---|
| Content libraries Coursera for Business, Udemy Business |
Licensed generic content; no live teaching; certificates without evidence | Generates organisation-specific courses and produces decision-grade evidence |
| LMS platforms Moodle, TalentLMS |
Administration shells that still require humans to make and teach the content | Is the trainer, the content team, and the examiner |
| AI tutor products | Tutor-only scope; no live voice classroom, no gated employment-grade assessment, no Indian-language delivery | Socratic voice examination with progression gates and a defensible evidence trail |
Protocol stage: analyse
The lab’s method runs in six rungs. Elenchium has reached analyse; everything past it is shown as not yet earned, which holds the project to the same standard as the registry’s empty slots.
- OBSERVE
- EXPERIMENT
- ANALYSEREACHED
- PROTOTYPEIN BUILD
- VALIDATESOFT PILOT · JAN 2027
- SCALELAUNCH · FEB 2027
Release scope
PHASE 1 · COURSE FACTORY
Thirteen-agent generation pipeline — curriculum planning, lesson authoring, deck construction, practice sets, per-candidate unique exams, QA gates, publication.
SOFT PILOT · JAN 2027
A live pilot cohort exercising the full evidence-packet loop end to end, ahead of commercial launch.
COMMERCIAL LAUNCH · FEB 2027
External enterprise and institutional sale, on the strength of pilot evidence rather than promise.
Market opportunity
| Layer | Definition | Estimate |
|---|---|---|
| TAM | India corporate training and skilling spend | ₹75,000+ Cr / yr |
| SAM | Technical onboarding, placement-readiness and role-certification programmes deliverable by an autonomous voice-first platform (English + Indian-language) | ₹6,000–9,000 Cr / yr |
| SOM 36 months |
Tamil Nadu and Chennai-metro engineering colleges plus SME/mid-market technology employers reachable by direct founder sales | ₹40–60 Cr / yr |
Planning estimates for sizing conversations, not audited market research.
- Model economics have crossed a threshold — course generation costs have fallen by orders of magnitude against a traditional content-development contract
- Indian-language voice AI reached production quality for Tamil–English code-mixed teaching, opening cohorts English-only products structurally cannot serve
- Employers increasingly demand evidence, not certificates — and the evidence packet is exactly the gap incumbents haven’t closed
Stack and infrastructure
- Next.js and Supabase on the product plane
- FastAPI and LangGraph orchestrating the agent pipeline
- Postgres for the system of record
- A three-layer memory stack — Mem0 for working memory, Qdrant for vector retrieval, and a Neo4j concept graph connecting curriculum objects
- Self-hosted LiveKit for the live voice classroom
- Sarvam Saaras and Bulbul for Tamil–English code-mixed speech
Positioning: where the human decides
The human stays exactly where a human is legally and ethically required: the employment decision. Elenchium generates, teaches, examines and documents — assembling the full evidentiary record — but it never decides whether someone is hired, confirmed, promoted, or held back.