Alethos Technologies (OPC) Private Limited · Pune, India
Alethos builds AI software that shows its evidence.
Every claim our products make can be traced back to the data behind it.
01
·The idea
Traceability is the organising principle.
The failure mode of an AI tool is not a wrong answer. It is a confident answer with nothing underneath it — no figures, no source, no period. It reads as finished work, so it is either used unchecked or checked again from scratch. If it has to be checked from scratch, the tool has saved nobody any time.
We build the other way round. An output is not prose with citations attached afterwards; it is assembled from figures that were extracted, recorded and recomputed, and the software can show that trail for any claim it makes. Where the trail cannot be produced, the claim is not made.
02
·See it work
Open the claim, see what it rests on.
This is the interaction the product is built around, shown here on its own.
03
·Our first product
StockSense AI Pro
StockSense AI Pro reads public company filings and market data and returns a structured research note. It is additive to the work an investment professional already does — it handles the reading, the extraction and the arithmetic, and shows the source for every figure it reports.
The note is a document, not a verdict. What it contains is set out beside this, section by section, and each figure in it opens the way the claim above does.
Access is by request while the product is in early release.
Report structure
| No. | Section | Contents |
|---|---|---|
| 01 | Business overview | Segments, how revenue is earned, and what changed this period |
| 02 | Financial summary | Revenue, margin and cash-flow series, each figure tied to its filing |
| 03 | Segment analysis | Reported segment performance, period over period |
| 04 | Balance sheet and cash flow | Debt levels, working capital and coverage, computed from filed statements |
| 05 | Filings timeline | What was filed, when, and what differs from the previous period |
| 06 | Evidence appendix | Every figure in the note with its source document and page |
04
·How we build
Three things we hold to in the engineering.
01
Bounded context per agent
Retrieval, extraction and verification run as separate agents, each given a narrow and explicit context. Narrow inputs are what make a step checkable on its own, and what stops a mistake in one stage from carrying silently into the next.
02
Validation before anything ships
Model output passes deterministic checks before it reaches a report: ratios are recomputed from the extracted values and totals are reconciled against the filed statement. Anything that does not reconcile is held back rather than printed. Changes run against a held-out evaluation set, and a regression blocks the release.
03
Cost engineered for production volume
Model routing per task, caching at the document level, and a measured token budget per report. We size the system against the real workload rather than a demonstration, because a pipeline that only holds at demo volume is not finished.
05
·Company
Alethos Technologies (OPC) Private Limited
StockSense AI Pro is our first product. Access requests and company enquiries go to the same address.
Alethos, from the Greek aletheia: unconcealment — things shown as they are.