The whole programme as one register: what each unit delivers, which of the eight working groups owns it, what “done” means, and what it waits on. Nothing here is a summary — it is the working document, published as it stands.
Each unit is owned by one group, so there is a single convenor accountable for it. Assignment follows the track, not the individual unit, which means new work inherits a group instead of waiting to be allocated. The Tech Team is not a ninth group — it carries the units whose deliverable is software.
A unit depends only on work in the same or an earlier phase. Where the real connection points forward it is recorded as a relation instead — open What done means on any row to see it. Nothing in a later phase is ever a blocker for an earlier one.
The Issue column links to the tracker. A green number with a tick is closed work — the record of a unit being finished. A dash means the unit is not on the board yet. 137 of 160 units are linked. The repository is private while the programme is in Phase 0–1, so these open for team members only.
| Unit | Issue | Title and definition of done | Owner | Phase | Status |
|---|---|---|---|---|---|
| SVU-001 | #4 | Approve the corpus database design and create it
What done means23 tables / 6 migrations applied to a shared database; loader writes live.
|
Tech Tech Team Lead | P0 | Blocked |
| SVU-002 | #5 | Add the dictionary tables to the corpus database
What done meanskosha_source, kosha_entry, pada_classification created; the 5-value word-class list in place. Migration 007 is written and awaiting sign-off.
|
Tech Tech Team Lead | P0 | Blocked |
| SVU-003 | #6 ✓ | Commission the two computing servers
What done means2 nodes racked, Ubuntu + CUDA + Docker + K8s installed, burn-in passed.
|
Tech Tech Team Lead | P0 | Done |
| SVU-004 | #7 ✓ | Install the databases and search services on the servers
What done meansPostgres, Qdrant, Neo4j, Typesense, MinIO, Keycloak up and reachable.
|
Tech Tech Team Lead | P0 | Done |
| SVU-005 | #8 | Constitute the scholarly review panel
What done means4–6 senior VVV/SVAMI scholars named as Tier 1/2 validators, terms agreed.
|
Tech Tech Team Lead | P0 | Not started |
| SVU-006 | #9 | Declare the licence for each of our three own works
What done meansSeparate licences declared for the generated Pada Kośa, the converted mūla, and the glosses we serve.
|
Tech Tech Team Lead | P0 | Blocked |
| SVU-007 | #10 | Set the code repository's public access and safeguards
What done meansPublic at launch; main branch protected; outside contributions need approval before their checks run; secret scanning on. Read access granted to the group convenors before then, so the issue board is usable by the people it assigns work to — the repository is private until Phase 1, which currently locks them out.
|
Tech Tech Team Lead | P0 | Not started |
| SVU-008 | #11 ✓ | Number every document and keep a register of them
What done meansEvery docs/ file numbered SVD-###; ledger + generated index; --check gate.
|
Tech Tech Team Lead | P0 | Done |
| SVU-009 | #12 ✓ | Register every text we hold, with its loading status
What done means280 distinct texts, stable IDs, load state derived from the bucket. The one register that new texts and new references extend without restructuring.
|
G1 VKG | P0 | Done |
| SVU-010 | #13 ✓ | Match the books the 72 volumes need against what we hold
What done means461 placements, 390 works, Work IDs, gap list.
|
G1 VKG | P0 | Done |
| SVU-011 | #14 ✓ | Generate project documents reproducibly, and reject invalid files
What done meansGenerators reproducible; schema-order audit passes; abort on invalid output.
|
Tech Tech Team Lead | P0 | Done |
| SVU-012 | #15 | Keep an off-site copy of the irreplaceable sources
What done means72 VVV volumes, Swadharma list and 2020 Bhāṣya Pilot replicated off the single workstation. Assessment risk R1, rated Critical.
|
G1 VKG | P0 | Not started |
| SVU-013 | — | Define what each group hands to the next, and when
What done meansThe workflow runs Collection → Processing and Ingestion → Pada Kośa and Specialist Analysis → Vedavākya Categorization → Testing → Review → Public and Multimedia. For each of those six boundaries: what is handed over, in what form, and what the receiving group may reject it for. Written down, because eight groups working in sequence stall at the boundaries, not in the middle. |
Tech Tech Team Lead | P1 | Ready |
| SVU-014 | — | Bring in outside funding through non-profits and matching grants
What done meansExternal funding onboarded via non-profit routes and corporate matching-grant programmes, to carry machine maintenance, infrastructure additions and scholarly contributions. Current work runs on internal resources alone.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-020 | #16 ✓ | Count every special VijayaDV character in the corpus
What done means32 codepoints U+E300–E352, 17,281 occurrences, counted from the pilot corpus.
|
G2 Chandra Chaganti | P1 | Done |
| SVU-021 | #17 ✓ | Sort the VijayaDV special characters into accent, marker and sandhi
What done meansAccent / structural marker / sandhi-āgama separated; ornament misreading corrected.
|
G2 Chandra Chaganti | P1 | Done |
| SVU-022 | #18 | Decide the standard character each remaining accent mark maps to
What done meansTargets fixed for dīrgha-svarita, adhyardha-svarita, kampa, praśleṣa, āgama; Sāma svara identities assigned.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-023 | #19 | Confirm that position-variant accent glyphs mean one accent
What done meansPosition-variant glyphs collapse to one logical combining mark; mapping rebuilt on it.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-024 | #20 | Convert the mūla of all four Vedas to standard characters
What done meansMapper run over the whole accented mūla; output byte-verified.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-025 | #21 | Prove every converted file converts back unchanged
What done meansEvery converted file round-trips byte-identical; gate wired into test_build.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-026 | #22 | Load the 120 files that are already machine-readable
What done meansNative DOCX/XLSX/TXT ingested; 71 MB; no OCR, no conversion.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-027 | #3 ✓ | Assemble the volumes already put through OCR
What done meansPaid-for Document AI output assembled from JSON shards to DOCX and uploaded. DONE 2026-08-02: 31 volumes, 11,824 pages, 0 failures, every one verified byte-exact against its source text before upload. ocr-documents/ now holds 68.
|
G2 Chandra Chaganti | P1 | Done |
| SVU-028 | #23 | Check each un-OCR'd PDF for existing text before paying for OCR
What done meansEach of the 34 classified text-layer vs genuine scan before any OCR spend. Two January jobs cost 530 pages of avoidable recognition for want of this gate. DONE 2026-08-08: all 34 classified, 9,539 pages assessed — 27 genuine scans, 4 with a usable text layer, 3 written in legacy fonts. |
G2 Chandra Chaganti | P1 | Done |
| SVU-029 | #24 | Run OCR on the books that are genuine scans
What done meansThe 27 files / 8,609 pages that gate O1 found to be genuine scans, put through the seven-stage pipeline. Every book registered in the digitisation ledger as it enters, so no page is ever recognised twice.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-030 | #25 | Extract the text from PDFs that already carry it
What done meansThe 4 files / 784 pages with a usable text layer extracted without OCR.
|
G2 Chandra Chaganti | P1 | Ready |
| SVU-031 | #26 | Convert the two old-format Kāṇva Saṃhitā files
What done meansThe .doc and .pub files converted to DOCX. DONE 2026-08-08: both converted through Office over COM, the only readers that render these formats faithfully.
|
G2 Chandra Chaganti | P1 | Done |
| SVU-032 | #27 | Archive the 5 PDFs whose DOCX we already hold
What done meansPDFs whose DOCX of the same work is already held: archived, not loaded. The move manifest is built; execution waits on the bucket-versioning decision.
|
G2 Chandra Chaganti | P1 | In progress |
| SVU-033 | #28 | Load the Kalpa Sūtra collection, split at sūtra level
What done means11 Śrauta + 8 Gṛhya Sūtra DOCX segmented at sūtra level.
|
G1 VKG | P1 | Blocked |
| SVU-034 | #29 | List every work quoted in the 72 volumes
What done meansAll 72 VVV volumes scanned; quoted works compiled and matched against holdings.
|
G1 VKG | P1 | Ready |
| SVU-035 | #30 | Put the 303 missing works in the order we should obtain them
What done meansGap list ordered by volume demand and licence status, with a source named per item. DONE 2026-08-08: 303 works scored and ordered.
|
G1 VKG | P1 | Done |
| SVU-036 | #31 | Bring the outside digital corpora into our own format
What done meansGRETIL, Muktabodha and VHP normalised to the locus scheme and PUA-safe encoding.
|
G1 VKG | P1 | Not started |
| SVU-037 | #32 | Mark where each ṛk, sūtra and śloka begins and ends
What done meansṚk / Sūtra / Śloka boundaries tagged across Tier 1.
|
G1 VKG | P1 | Not started |
| SVU-038 | #33 | Freeze the numbering that addresses every sentence in the corpus
What done meansFixed-width zero-padded śākhā·kāṇḍa·praśna·anuvāka·mantra key, widths set against the master register with one digit headroom. Must precede any layer tree. The eight VedaNidhi vākya fields are frozen; the pada layer is L9. |
Tech Tech Team Lead | P1 | Blocked |
| SVU-039 | #34 | Index every passage for meaning-based search
What done meansAll segments indexed for similarity search (multilingual-e5-large or LaBSE).
|
Tech Tech Team Lead | P1 | Not started |
| SVU-040 | #35 | Link ṛk to devatā, sūkta to ṛṣi, mantra to rite
What done meansThe relation graph loaded from the Swadharma taxonomy.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-041 | #36 | Load the four 2020 Bhāṣya Pilot volumes as the gold standard
What done meansThe four DOCX ingested as unit/pada/meaning/bhashya. MAP is Mūla-Artha-Pratipatti: the mūla, its meaning, and how that meaning is arrived at.
|
G2 Chandra Chaganti | P1 | Blocked |
| SVU-042 | #37 | Resolve every grammatical and śāstric citation in the Bhāṣya
What done meansCommentary processed so grammatical and śāstric citations resolve to targets.
|
G2 Chandra Chaganti | P1 | Not started |
| SVU-043 | #38 | Group each mantra and brāhmaṇa passage with its Padapāṭha and Bhāṣyam
What done meansEach sentence-group carries its Padapāṭha and Bhāṣyam together.
|
G2 Chandra Chaganti | P1 | Not started |
| SVU-044 | #39 | Compute meaning, analysis and presentation for a mantra on demand
What done meansMeaning–Analysis–Presentation computed on demand, not stored per form.
|
G2 G2 → Tech | P1 | Not started |
| SVU-044.1 | #114 | Spec — MAP analysis contract — what the reader is shown and from what
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G2 Chandra Chaganti | P1 | Ready |
| SVU-044.2 | #115 | Build — Compute meaning, analysis and presentation for a mantra on demand
What done meansMeaning–Analysis–Presentation computed on demand, not stored per form.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-046 | #41 | Publish finished material to vaakya.vedanidhi.in as it is ready
What done meansProcessed material published incrementally rather than held to the end.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-048 | #43 | File the 33 documents left in the staging folder
What done meansMoved from uploads/ into Shrouta/ and Shaastra/ with correct naming. The manifest is built — 19 firm, 7 needing a scholar's ruling, 7 held for want of a prefix or a licence call. Execution waits on the bucket-versioning decision, because a move in a bucket without versioning is an irreversible delete.
|
G1 VKG | P1 | In progress |
| SVU-049 | — | Recover the text from PDFs written in legacy fonts
What done meansThe 3 files / 146 pages whose text layer is a legacy non-Unicode font: the font's encoding identified and mapped, so the text is recovered without OCR. Gate O1 found this third class, which SVD-008 did not anticipate — the files are neither genuine scans nor plainly extractable.
|
G2 Chandra Chaganti | P1 | Ready |
| SVU-050 | #44 | Load the 62 dictionaries and make their words searchable
What done means62 kośas / 1.79 M rows loaded; 312,682 distinct words queryable.
|
G3 VKG | P1 | Blocked |
| SVU-051 | #45 ✓ | Publish the census of what the dictionaries contain
What done meansKośa-gata Padāni issued with 11 decisions and reproducible figures.
|
G3 VKG | P1 | Done |
| SVU-052 | #46 | Decide what may be shown from the 12 copyrighted dictionaries
What done meansDecision recorded and enforced in code before any public reader release.
|
G3 VKG | P1 | Blocked |
| SVU-053 | #47 | Fix which dictionary the reader is shown first
What done meansOrder fixed: K1 tradition → Mahākośa → MW → remainder by date.
|
G3 VKG | P1 | Blocked |
| SVU-054 | #48 | Flag the 106,014 words attested in only one dictionary
What done meansAdmitted with an uncorroborated flag rather than silently or not at all.
|
G3 VKG | P1 | Blocked |
| SVU-055 | #49 | Load every Padapāṭha we hold, with its accents intact
What done meansEvery held padapāṭha loaded with svara intact into `pada` + `svara_mark`, and peeled uniformly at load time: the Iti part separated from the Veṣṭana Padam, Upasarga and Pragṛha to yield the Prayoga Padam, with the verbatim form kept beside it for recitation. Both forms stored, so the Krama and Vikṛti layers build on the peeled padas. Needs migrations 001-002 applied and nothing more — NOT migration 007, and NOT the character mapping: pada_devanagari is PUA-safe and stores text verbatim, so VijayaDV glyphs load byte-exact today. |
G3 VKG | P1 | Blocked |
| SVU-057 | #51 | Generate a Padapāṭha for texts that lack one
What done meansRule-generated padapāṭha for Veda texts lacking one, scholar-validated.
|
G3 G3 → Tech | P1 | Not started |
| SVU-057.1 | #116 | Spec — Rules for generating Pada-pāṭha where none is attested
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P1 | Ready |
| SVU-057.2 | #117 | Build — Generate a Padapāṭha for texts that lack one
What done meansRule-generated padapāṭha for Veda texts lacking one, scholar-validated.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-058 | #52 | Build the Vedic lexicon: words and meanings drawn from the Bhāṣya
What done meansUnique words with meanings extracted from Veda Bhāṣyam; the Vedic lexicon.
|
G3 G3 → Tech | P1 | Not started |
| SVU-058.1 | #118 | Spec — Method for extracting the Vaidika lexicon from Bhāṣya
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P1 | Ready |
| SVU-058.2 | #119 | Build — Build the Vedic lexicon: words and meanings drawn from the Bhāṣya
What done meansUnique words with meanings extracted from Veda Bhāṣyam; the Vedic lexicon.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-059 | #53 ✓ | Derive words from dhātu and pratyaya, with the sūtra chain
What done means4,955 forms / 40 dhātus / 1.08 s / 0 errors; 73 tests; full sūtra chain per form.
|
G3 VKG | P1 | Done |
| SVU-060 | #54 | Fix how accents are printed in published forms
What done meansOne line in svara.py settled; every published accented form depends on it.
|
G3 VKG | P1 | Blocked |
| SVU-061 | #55 | Decide whether pracaya and ekaśruti are applied
What done meansApplied, not applied, or reader-side toggle. ~10% of forms affected.
|
G3 VKG | P1 | Blocked |
| SVU-062 | #56 | Measure our derived accents against the gold corpus, vowel by vowel
What done meansDerived accent compared vowel-by-vowel with the 2020 Bhāṣya Pilot; agreement rate reported with a named disagreement list.
|
G3 G3 → Tech | P1 | Blocked |
| SVU-062.1 | #120 | Spec — Accent-agreement test design and acceptance threshold
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P1 | Blocked |
| SVU-062.2 | #121 | Build — Measure our derived accents against the gold corpus, vowel by vowel
What done meansDerived accent compared vowel-by-vowel with the 2020 Bhāṣya Pilot; agreement rate reported with a named disagreement list.
|
Tech Tech Team Lead | P1 | Not started |
| SVU-063 | #57 | Decide how many derived forms we generate
What done means |
G3 VKG | P1 | Blocked |
| SVU-064 | #58 | Generate the full set of forms at the agreed size
What done meansComplete generation at the agreed target; shards + provenance report.
|
G3 VKG | P1 | Blocked |
| SVU-065 | #59 | Build the fast word-lookup index
What done meansvidyut-kosha index built for instant lookup at scale; Rust toolchain provisioned.
|
G3 VKG | P1 | Blocked |
| SVU-066 | #60 | Write the ~3,400 morpheme meanings in Telugu, Kannada, Hindi and English
What done means~3,400 authored glosses × four languages. Telugu leads — the team's native expertise gives the subtlest sense.
|
G3 VKG | P1 | Not started |
| SVU-067 | #61 | Give the nirvacana of a word, with competing etymologies
What done meansNirvacana for any word with competing etymologies cited beside the Pāṇinian vyutpatti. Requires acquiring a Nighaṇṭu and the Nirukta — absent today.
|
G3 G3 → Tech | P2 | Blocked |
| SVU-067.1 | #122 | Spec — Nirvacana model — sources, competing etymologies, presentation
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P2 | Ready |
| SVU-067.2 | #123 | Build — Give the nirvacana of a word, with competing etymologies
What done meansNirvacana for any word with competing etymologies cited beside the Pāṇinian vyutpatti. Requires acquiring a Nighaṇṭu and the Nirukta — absent today.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-068 | #62 | Establish the liṅga of each stem from the liṅga authorities
What done meansNāma-liṅgānuśāsana with Mahābhāṣya and Kāśikā establishing liṅga per stem.
|
G3 VKG | P2 | Not started |
| SVU-069 | #63 | Apply sandhi by rule, with the accent change at each junctura
What done meansRule-verified sandhi with svara transformation at every junctura.
|
G3 G3 → Tech | P2 | Not started |
| SVU-069.1 | #124 | Spec — Sandhi rule inventory and svara behaviour at each junctura
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P2 | Ready |
| SVU-069.2 | #125 | Build — Apply sandhi by rule, with the accent change at each junctura
What done meansRule-verified sandhi with svara transformation at every junctura.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-070 | #64 | Generate the Krama pāṭha of any Saṃhitā passage
What done meansKrama pāṭha for any Saṃhitā segment, rule-verified.
|
G3 G3 → Tech | P2 | Not started |
| SVU-070.1 | #126 | Spec — Krama construction rules
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P2 | Ready |
| SVU-070.2 | #127 | Build — Generate the Krama pāṭha of any Saṃhitā passage
What done meansKrama pāṭha for any Saṃhitā segment, rule-verified.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-071 | #65 | Generate the eight Vikṛti pāṭhas, with accents preserved
What done meansJaṭā, Mālā, Śikhā, Rekhā, Dhvaja, Daṇḍa, Ratha, Ghana — with svara preserved.
|
G3 G3 → Tech | P2 | Not started |
| SVU-071.1 | #128 | Spec — The eight Vikṛti forms — construction rules per form
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P2 | Ready |
| SVU-071.2 | #129 | Build — Generate the eight Vikṛti pāṭhas, with accents preserved
What done meansJaṭā, Mālā, Śikhā, Rekhā, Dhvaja, Daṇḍa, Ratha, Ghana — with svara preserved.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-072 | #66 ✓ | Varṇa Krama — already shipped, so settle reuse terms only
What done meansAkṣara-by-akṣara decomposition with svara and mātrā from Śikṣā/Prātiśākhya. ALREADY SHIPPED as a spinoff app on Play Store and App Store — do not rebuild. Remaining work is reuse and citation terms, not construction. Settled with it: no adapter will be trained to imitate this engine; the model calls the engine, because approximating an exact algorithm introduces error where none existed. |
G3 VKG | P2 | Done |
| SVU-073 | #67 | Look up a form and return its analyses for the hover
What done meansForm → analyses; the grammar hover contract and the spell-check predicate.
|
G3 VKG | P2 | In progress |
| SVU-074 | #150 | Join a Veda word to its dictionary entry and its derivation
What done means`pada` has no lemma column, so a Veda word occurrence cannot be joined to its kośa attestation or its Pāṇinian derivation — which is the entire point of the Vaidika Pada Kosham. Add a `lemma` column on `pada` or a `pada_lemma` bridge, keyed to normalised SLP1. Propose alongside migration 007.
|
G3 VKG | P1 | Blocked |
| SVU-080 | #68 | Connect cross-references across Veda, Vedāṅga, Upāṅga and Upaveda
What done meansCross-references across the four verticals connected wiki-style.
|
G5 Vijay Krishna | P2 | Not started |
| SVU-081 | #69 | Translate mantras into Indian languages, grounded in the Bhāṣya
What done meansIndian-language translation of mantras derived solely from Bhāṣya; where none exists, Sāyaṇa's principles are followed explicitly.
|
G3 G3 → Tech | P2 | Not started |
| SVU-081.1 | #132 | Spec — Anuvāda method — grounding every rendering in Bhāṣya
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G3 VKG | P2 | Ready |
| SVU-081.2 | #133 | Build — Translate mantras into Indian languages, grounded in the Bhāṣya
What done meansIndian-language translation of mantras derived solely from Bhāṣya; where none exists, Sāyaṇa's principles are followed explicitly.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-082 | #70 | Find passages similar in word and in meaning
What done meansSimilarity across words, meanings and contexts.
|
G5 G5 → Tech | P2 | Not started |
| SVU-082.1 | #134 | Spec — Samānatā — what counts as similarity in śabda and in artha
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G5 Vijay Krishna | P2 | Ready |
| SVU-082.2 | #135 | Build — Find passages similar in word and in meaning
What done meansSimilarity across words, meanings and contexts.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-083 | #71 | Analyse a sentence across all the verticals
What done means |
G5 Vijay Krishna | P2 | Not started |
| SVU-084 | #72 | Assemble the instruction-and-answer set that trains the model
What done means~500K instruction–response pairs from Tier 1 + Tier 2.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-085 | #73 | Fine-tune the base model on the Vedic corpus
What done means4-bit adapter training, 3 epochs, on a scholar-graded benchmark. Model choice is settled by what one node's 24 GB will hold — no 70B.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-086 | #74 | Answer from retrieved passages rather than from memory
What done meansQuestion → passage retrieval → grounded answer; 200-question benchmark passed.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-087 | #75 | Improve answers from scholars' ratings of paired outputs
What done means500 scholar-rated output pairs; reward model; one preference-tuning pass.
|
Tech Tech Team Lead | P2 | Not started |
| SVU-088 | #76 | Re-measure training time on the servers we actually have
What done meansThe 10–14 days/epoch figure re-measured on delivered hardware before Phase-2 dates are committed externally.
|
Tech Tech Team Lead | P2 | Blocked |
| SVU-089 | #77 | Generate a complete Śrauta rite sequence, every step cited
What done meansComplete rite sequences — ṛtvij, mantra with svara, havirdravya, kāla-nirṇaya — every step carrying a traceable sūtra citation.
|
G5 G5 → Tech | P3 | Not started |
| SVU-089.1 | #136 | Spec — Śrauta Prayoga sequence model and citation requirements
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G5 Vijay Krishna | P3 | Ready |
| SVU-089.2 | #137 | Build — Generate a complete Śrauta rite sequence, every step cited
What done meansComplete rite sequences — ṛtvij, mantra with svara, havirdravya, kāla-nirṇaya — every step carrying a traceable sūtra citation.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-090 | #78 | Generate Sāmaveda stotras with viṣṭuti and stobha
What done meansStotriyā assembly, viṣṭuti patterns, stobha insertion; numeric-svara and sargam.
|
G4 G4 → Tech | P3 | Not started |
| SVU-090.1 | #138 | Spec — Stotriyā assembly, viṣṭuti patterns and stobha rules
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G4 Natraj Kavuri | P3 | Ready |
| SVU-090.2 | #139 | Build — Generate Sāmaveda stotras with viṣṭuti and stobha
What done meansStotriyā assembly, viṣṭuti patterns, stobha insertion; numeric-svara and sargam.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-091 | #79 | Teach the model the Vikṛti pāṭhas from rule-generated forms
What done meansAdapter trained on rule-generated forms for 10,000 seed Ṛk units.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-092 | #80 | Analyse arguments in Tarka form: pañcāvayava and Navya-Nyāya
What done meansPañcāvayava syllogisms, pūrvapakṣa–siddhānta frames, Navya-Nyāya analysis; śābda-bodha style with pariṣkāra.
|
G5 G5 → Tech | P3 | Not started |
| SVU-092.1 | #140 | Spec — Śābda-bodha and pariṣkāra representation for Tarka
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G5 Vijay Krishna | P3 | Ready |
| SVU-092.2 | #141 | Build — Analyse arguments in Tarka form: pañcāvayava and Navya-Nyāya
What done meansPañcāvayava syllogisms, pūrvapakṣa–siddhānta frames, Navya-Nyāya analysis; śābda-bodha style with pariṣkāra.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-094 | #82 | Analyse a recitation's sound: svara, modulation and volume
What done meansAudio in, notation out: pitch and svara extraction from recorded Sāmagāna, with voice modulation, decibel measurement and the other speech characteristics reported alongside.
|
G4 G4 → Tech | P3 | Not started |
| SVU-094.1 | #142 | Spec — Svara extraction and notation conventions for Sāmagāna
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G4 Natraj Kavuri | P3 | Ready |
| SVU-094.2 | #143 | Build — Analyse a recitation's sound: svara, modulation and volume
What done meansAudio in, notation out: pitch and svara extraction from recorded Sāmagāna, with voice modulation, decibel measurement and the other speech characteristics reported alongside.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-095 | #83 | Re-frame a melody onto a new chandas
What done meansŪha: melody re-framed onto new chandas, learned from Grāmageya–Ūha pairs.
|
G4 G4 → Tech | P4 | Not started |
| SVU-095.1 | #144 | Spec — Ūha — how melody is re-framed onto new chandas
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G4 Natraj Kavuri | P4 | Ready |
| SVU-095.2 | #145 | Build — Re-frame a melody onto a new chandas
What done meansŪha: melody re-framed onto new chandas, learned from Grāmageya–Ūha pairs.
|
Tech Tech Team Lead | P4 | Not started |
| SVU-097 | #85 | Interpret a passage through four darśanas side by side
What done meansFour lens adapters — Tarka, Vyākaraṇa, Mīmāṃsā, Vedānta — each with retrieval restricted to its own corpus; four-column comparative view.
|
G5 G5 → Tech | P3 | Not started |
| SVU-097.1 | #148 | Spec — The four lenses — scope and retrieval boundary of each
What done meansA written specification a builder can implement without further scholarly input: rules, worked examples, edge cases, and the acceptance test. Reviewed before build starts.
|
G5 Vijay Krishna | P3 | Ready |
| SVU-097.2 | #149 | Build — Interpret a passage through four darśanas side by side
What done meansFour lens adapters — Tarka, Vyākaraṇa, Mīmāṃsā, Vedānta — each with retrieval restricted to its own corpus; four-column comparative view.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-098 | #86 | Show a passage in five layers, mūla to related passages
What done meansMūla/IAST → Padapāṭha → Bhāṣya → grammar hover → related-passage sidebar.
|
Tech Tech Team Lead | P3 | In progress |
| SVU-099 | #87 | Show every word's grammar on hover, by lookup not guesswork
What done meansEvery word hoverable: dhātu, pratyaya, vibhakti, vacana, liṅga, sūtra ref. No model at read time — deterministic lookup only.
|
Tech Tech Team Lead | P3 | In progress |
| SVU-100 | #88 | Rite generator for scholars, with its citation trail
What done meansScholar-tier Prayoga Mode with citation trail.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-101 | #89 | Screen that turns a Saṃhitā passage into all eight Vikṛti forms
What done meansAny Saṃhitā segment → all eight forms with sandhi and svara shown.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-102 | #90 | Sāma Studio: stotra generator, notation viewer, sound panel
What done meansStotra generator, notation viewer, sound-analysis panel.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-103 | #91 | Tarka workbench, and the akṣara-by-akṣara Varṇa Krama view
What done meansBhāṣya passage parsed into formal argument structure; akṣara decomposition view served by the shipped Varṇa Krama engine.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-104 | #92 | Stotra library by devatā and chandas, and an etymology explorer
What done meansStotra corpus filterable by devatā, kartā, chandas, occasion; Nirukta explorer.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-105 | #93 | Notes connecting a passage to science, philosophy and the arts
What done meansConnections to science, philosophy and the arts surfaced per passage.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-106 | #94 | Serve the model on the second node, with caching
What done meansInference deployed on Node B; common queries cached; requests batched.
|
Tech Tech Team Lead | P3 | Not started |
| SVU-107 | #95 | Run the pilot with 25–50 faculty and scholars
What done means25–50 SVAMI faculty, VVV scholars and researchers using the system.
|
G7 Vishwanath | P4 | Not started |
| SVU-108 | #96 | Score answers on all 2,000 Swadharma topics against scholar answers
What done meansAll 2,000 topics scored against scholar reference answers on a 5-point scale.
|
G6 Gayathri Ramasubramanian | P4 | Not started |
| SVU-109 | #97 | Check generated rites against the 2026 Agniṣṭoma performance
What done meansGenerated sequences checked against an actual Śrauta performance.
|
G7 Vishwanath | P4 | Not started |
| SVU-160 | — | Improve the speech-to-text model on Vedic recitation
What done meansThe IISc Bangalore model is open source, runs on the project GPUs today and is already above 80% accurate. Fine-tuned in house against recitation until it reaches the 99% target, with accuracy reported per śākhā rather than as one average — a model that is excellent on one śākhā and poor on another is not a 99% model.
|
G4 Natraj Kavuri | P1 | In progress |
| SVU-161 | — | Record volunteers chanting each śākhā, with correct intonation
What done meansAn audio corpus the project owns outright: volunteers recording across the śākhās, with authentic svara. This is the constraint that limits everything else in this track — the recognition model today learns from pre-trained outside datasets, not from project-created audio, and there is no large corpus of Vedic recitation to draw on. Recording is how that changes. |
G4 Natraj Kavuri | P1 | Ready |
| SVU-162 | — | Speak Vedic text aloud with the svara correct
What done meansText to speech that carries svara, not merely phonemes. Judged by reciters: a rendering with correct phonemes and wrong svara is a failure, not a near miss.
|
G4 Natraj Kavuri | P2 | Not started |
| SVU-163 | — | Settle the licence and access terms for the training audio
What done meansWhich recitation audio may be used for training, on what terms, and who may reach the model and the corpus. Two things force this: the outside datasets behind the current model carry licence constraints, and access is at present held open only by specialist logins rather than by a stated policy. The audio the project records itself (SVU-161) is the clean route, and needs its own consent terms from the reciters.
|
G4 Natraj Kavuri | P1 | Ready |
| SVU-170 | — | Test every group's output for function and for content
What done meansEach group's output tested on both counts — does it work, and is it right — with findings returned to the originating group as specific, reproducible defects rather than impressions. Coverage reported per group, so a group whose work is not being tested is visible.
|
G6 Gayathri Ramasubramanian | P1 | Ready |
| SVU-171 | — | Sample scholarly quality: random at first, then systematic
What done meansHuman scholarly checking of processed output, beginning as random sampling to find where the errors are, then becoming systematic on the passages and categories where sampling showed the error rate is high. The sampling rate and the observed error rate are both reported — a quality claim with neither is not a measurement.
|
G6 Gayathri Ramasubramanian | P1 | Ready |
| SVU-175 | — | Final review before anything reaches the public
What done meansThe last gate: processed and tested content reviewed for scholarly quality and accuracy, and either passed or returned, before it moves to presentation or publication. Nothing reaches a public surface without passing here, and what passed is recorded, so a published error can be traced to the review that cleared it. |
G7 Vishwanath | P2 | Not started |
| SVU-140 | — | Approve the Mīmāṃsā sentence-classification vocabularies and record formats
What done meansThe controlled vocabularies and the three record formats frozen by the Editorial Board: the three top classes (Mantra / Brāhmaṇa / Nāmadheya), the Vidhi–Niṣedha–Arthavāda tree with both Vidhi axes, and the twelve-fold Mantra and Brāhmaṇa hover lists. Nothing is tagged before this is signed.
|
G5 Vijay Krishna | P1 | In progress |
| SVU-141 | — | Tag the Darśapūrṇamāsa pilot passage by hand
What done meansTS 2.5 and its Brāhmaṇa portion, about 300 sentences, tagged by scholars against the approved vocabularies. This is the calibration set: it fixes what the marks look like in practice before any rule is written.
|
G5 Vijay Krishna | P1 | Not started |
| SVU-142 | — | Tag sentences automatically from the visible marks in the text
What done meansFirst-pass classifier over tokenised text: vidhi-liṅ and loṭ endings, kṛtya suffixes, na/mā with the optative, the narrative particles vai/ha/hi, and nominal apposition for Nāmadheya. Every tag carries a confidence score. Sentences are grouped by ekavākyatā before the class is finally assigned. Agreement with the hand-tagged pilot is the acceptance test.
|
G5 Vijay Krishna | P2 | Not started |
| SVU-143 | — | Review console where a scholar confirms or overrides each automatic tag
What done meansEvery automatic tag reviewable, with the marks that produced it shown beside it; ambiguous sentences queued rather than guessed. Approve / amend / reject is recorded per sentence with the reviewer named.
|
G5 Vijay Krishna | P2 | Not started |
| SVU-144 | — | Enter the first Adhyāya of the Nyāyamālā as Adhikaraṇa records
What done meansEvery pāda of Adhyāya 1 entered in the five-limb structure — viṣaya, saṃśaya, pūrvapakṣa, siddhānta, and the resulting tag decision — anchored on the opening kārikā, which is stored verbatim. The subject sentences are stored as corpus references, not loose text, so each Adhikaraṇa links back to the tagged corpus. This adhyāya validates the template before the bulk run.
|
G5 Vijay Krishna | P1 | Not started |
| SVU-145 | — | Enter all the Nyāyamālā Adhikaraṇas, with Bhāṭṭa and Prābhākara positions
What done meansAbout 1,200 Adhikaraṇas across the twelve Adhyāyas and sixty Pādas, each with the Bhāṭṭa siddhānta and, where it diverges, the Prābhākara position with its mechanism and its texts. Board workflow: draft → reviewed → approved. |
G5 Vijay Krishna | P2 | Not started |
| SVU-146 | — | Registry of rites and episodes, so one sentence can be traced through every rite
What done meansA Ritual → Episode → Sub-episode registry, śākhā-neutral, beginning with Darśapūrṇamāsa, Agniṣṭoma and Cāturmāsya, and the tagging that records why a sentence appears in an episode: native to it, employed by viniyoga, present by transfer, present in modified form, or annulled there. From any sentence, every episode it serves; from any episode, every sentence it employs. |
G5 Vijay Krishna | P2 | Not started |
| SVU-147 | — | Three services: classify a sentence, explain its tag, walk the episode graph
What done meansClassify returns the tag with its confidence and the marks behind it; explain returns the Adhikaraṇas that govern the decision, both schools' positions included; episode-graph walks Layer IV. These three are what the reader's hover and the model both call. This is the Mīmāṃsā Vākyārtha engine, the tenth of the family. |
G5 Vijay Krishna | P3 | Not started |
| SVU-148 | — | Training set of tag decisions with the reasoning behind each
What done meansApproved records exported as (sentence → tag decision → reasoning chain) triples, so the model learns to justify a classification in adhikaraṇa form: viṣaya, saṃśaya, pūrvapakṣa, siddhānta. Each Adhikaraṇa also becomes one retrieval document, kārikā first. |
G5 Vijay Krishna | P3 | Not started |
| SVU-115 | #98 | Put this plan and its current status on sarvaveda.info
What done meansPublic site carries the finalised 4-stage plan and current status.
|
Tech Tech Team Lead | P5 | Ready |
| SVU-116 | #99 | Publish pages as fixed HTML with sitemaps, so search engines read them
What done meansContent pages pre-rendered at build time, with schema.org markup and sitemaps, so search engines see the corpus and passages surface in general search. The precedent site builds its pages in the browser and is invisible to search.
|
Tech Tech Team Lead | P5 | Not started |
| SVU-117 | #100 | Let a reader's correction on the page become a proposed change
What done meansReader correction opens a pull request against an edit-granular repository. This one mechanism produced 197 forks and 59 PRs for Aṣṭādhyāyī with no admin panel.
|
Tech Tech Team Lead | P5 | Blocked |
| SVU-118 | #101 | Open the site with General, Scholar and Institutional access
What done meansThe three access tiers live.
|
Tech Tech Team Lead | P5 | Not started |
| SVU-119 | #102 | Machine access for researchers, documented and rate-limited
What done meansVedic references with simple meanings, documented and rate-limited.
|
Tech Tech Team Lead | P5 | Not started |
| SVU-121 | #104 | Answer any topic from the śāstra, with sourced citations
What done meansAny topic answered from the śāstra with sourced citations, including societal references.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-122 | #105 | Production route from corpus to published video, audio and web
What done meansProduction route from corpus to published media.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-123 | #106 | Run regular scholarly webinars from the corpus
What done meansRegular scholarly webinars drawing on the corpus.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-124 | #107 | Generate the VVS textbook series from the validated corpus
What done meansTextbook series generated from the validated corpus.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-125 | #108 | Produce short videos at scale, each scholar-approved before release
What done meansShort-form output across thousands of topics, written for a young social-media audience and turning scholarly work into something a non-specialist follows. Every reel is vetted by the testing and review groups before it is published — the audience is wide, so an error travels.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-180 | — | Presentation that explains the initiative to a general audience
What done meansOne well-made public presentation of what SarvaVeda is and why it matters, in terms a non-specialist follows — the piece the outreach and funding conversations both start from.
|
G8 Chandra Chaganti | P5 | Ready |
| SVU-126 | #109 | Publish daily across the channels, including on current events
What done meansScheduled daily publication across channels, carrying timely śāstra commentary on contemporary events — grounded in citation, not opinion.
|
G8 Chandra Chaganti | P5 | Not started |
| SVU-128 | #111 | Publish the consultation papers with their licence settled
What done meansPapers and resource list public, with a documents licence and the restricted-kośa disclosure settled.
|
Tech Tech Team Lead | P5 | Blocked |
| SVU-129 | #112 | Retrain each quarter on new volumes and feedback
What done meansQuarterly fine-tuning incorporating new volumes, transcripts and feedback.
|
Tech Tech Team Lead | P5 | Not started |
| SVU-130 | #113 | Transcribe 1,000 hours of video and align it to the text
What done means1,000 hours through Whisper large-v3 with diarisation, aligned to text units.
|
G4 Natraj Kavuri | P5 | Not started |
| SVU-149 | — | Public Mīmāṃsā pages: corpus browser, tag explorer, Adhikaraṇa reader
What done meansThe corpus browsable with hover tool-tips carrying the twelve-fold Mantra and Brāhmaṇa vocabularies; a tag explorer; the Adhikaraṇa database readable with a Bhāṭṭa / Prābhākara toggle; episode traversal. Public read, annotator login for write. Tool-tips are an enhancement — the page stays readable without them.
|
G5 Vijay Krishna | P5 | Not started |
| SVU-150 | — | Reorganise vedavishtaram.in into seven top-level sections
What done meansA persistent top navigation with seven tabs — Home, About, VVS Series, Mīmāṃsā Framework, SarvaVeda LLM, Publications & Downloads, Contact — each a module landing page with its own sub-navigation. Every existing block re-homed, nothing deleted, and a redirect map from the old anchors with no broken inbound links. Site-wide standards apply: वेदविज्ञानविष्टरम् with the short a, SGSP as series publisher, IAST beside Devanāgarī.
|
Tech Tech Team Lead | P5 | Ready |
| SVU-151 | — | Build the VVS catalogue straight from the Master Register
What done meansDivision landing pages and volume cards generated from the Master Register export at build time — 214 entries, 28 category codes — so the register stays the single source of truth and the site cannot drift from it. Live content reconciled against the register during migration so nothing published is lost.
|
Tech Tech Team Lead | P5 | Not started |
Each was removed because its work is carried elsewhere. A retired number is never reused, so the record stays legible.