Skip to content

Research for useful intelligence.
Evidence you can inspect.

Indian-language models, tool-using agents and reusable capability. Explore what we’ve measured, how systems are designed and the questions still open.

Translucent lavender, apricot and sage forms joined by woven threads, an abstract illustration of connected research.

Language. Capability. Verification.

Three connected questions guide the work. Each needs a check that can disagree with us.

Across languages

How does behavior change between native script, Romanized input and unfamiliar phrasing?

Language evaluations

Beyond one answer

When can a useful capability be composed, reused or withdrawn without hiding its limits?

Explore NIF

Checks that matter

Can verification reject a convincing failure, rather than simply confirm a successful process?

Agent architecture

Read the work.
Keep the limits in view.

Measured results, architecture notes and research directions—clearly distinguished, with the underlying detail one click away.

Research report

Speech to text, no settings

An open speech model turned into a tool that finds the language itself, and why word error rate misleads in Indian languages.

Read the work

Research report

Reading 22 Indian scripts

A small open model from 36 to 76 on Sarvam’s Indic OCR bench: synthetic print, real proofread pages, and long pages that still end too soon.

Read the work

Research direction

Neural Intelligence Fabric

When should useful work become reusable capability? An investigation into composition, verification and reversibility.

Read the work

Measured note

Reusing verified answers

When a question returns in another language, a verified earlier answer matches thinking mode at about a fifth of its tokens.

Read the work

Inference engineering

A faster 4-bit layer

Where a 24B model spends its time while it writes, and a measured speed-up that keeps our adapters' 4-bit base exact.

Read the work

Evaluation report

Indian-language evaluation

Maths, knowledge, refusal and over-refusal beside the sarvam-m base model. Read the setup, saved results and limitations together.

Read the work

Architecture note

From proposal to checked operation

How generation, task state, tools and permissions fit together. Implemented mechanisms are not a guarantee of task success.

Read the work

Concept comparison

Different systems. Different jobs.

The distinct roles of models, retrieval, tool protocols and reusable capability. An architectural comparison, not a leaderboard.

Read the work

Historical report · Superseded

Earlier language experiments

The historical evaluation, with language slices, a corrected baseline and regressions preserved. Superseded by the current report.

Read the work

A result is a beginning.
Not the whole story.

A module can exist before its usefulness is established. A passing evaluation does not cover every real-world task.

Name the conditions

Model, baseline, task, sample counts and evaluation settings belong beside a result.

Keep the failures

Regressions, refusal tradeoffs and unsuccessful cases matter as much as improvements.

Leave room to disagree

Separate observations from interpretations and open hypotheses. State what remains untested.