docs: update daily report with Session 5 — LanceDB fix, ingestion, LLM switch, Rust modules

Author: uet-santx

uet_web/Docs_software_design/daily_reports/2026-03-20.md

@@ -284,3 +284,77 @@ POST /chat {"prompt": "UET คืออะไร"} → 200 OK
 8. ✅ Safe __init__.py imports
 9. ✅ API key security (.env, not hardcoded)
 10. ✅ End-to-end verified (server starts, LLM responds, memory works)
+
+---
+
+## Session 5: LanceDB Fix, Knowledge Ingestion, LLM Switch, Rust Modules (same day)
+
+### LanceDB Vector Schema Fix
+
+The `OmegaSearch` engine was failing with:
+```
+RuntimeError: Data type is not a vector (FixedSizeListArray), but Float64
+```
+
+**Root cause**: Vectors stored as `list<double>` (variable-length) instead of `fixed_size_list<float>[N]` (fixed-length) that LanceDB requires for ANN search.
+
+**Fix**: Created `research_uet/knowledge_base/migrate_lance_schema.py` that:
+1. Reads all 5609 rows from old table
+2. Builds explicit PyArrow schema with `pa.list_(pa.float32(), N)`
+3. Drops old table, creates new with proper vector types
+4. Verifies vector search works
+
+Result: `semantic_vec: fixed_size_list<float>[8]`, `uet_vec: fixed_size_list<float>[20]`
+
+### Knowledge Ingestion
+
+Created `uet_agents/ingest_docs.py` — bulk ingestion script that reads all markdown docs and feeds them to the Semantic Engine.
+
+- **116 documents** from `research_uet/Docs/` and `research_uet/Doc/`
+- **5984 chunks**, **18085 vocab terms** (up from 7 test chunks)
+- Supports both API mode and direct mode
+
+### LLM Model Switch
+
+Switched from `qwen/qwen-2.5-7b-instruct` (mixed Chinese mid-response) to `z-ai/glm-4.7-flash`:
+- Responds in Thai consistently ✅
+- References source data accurately ✅
+- Handles reasoning model format (`content` + `reasoning` fields)
+- $0.06/M input tokens (cheap)
+
+Also fixed: `config.toml` with API keys was tracked in Git → untracked and added to `.gitignore`
+
+### Rust Modules Integration
+
+Added 4 under-development Rust modules to workspace:
+
+| Module | Files | Status |
+|--------|-------|--------|
+| `uet_economic` | 6 source files (difficulty, issuance, portfolio) | ✅ Compiles |
+| `uet_governance` | 6 source files (voting, proposals, policy) | ✅ Compiles |
+| `uet_market` | 7 source files (AMM, price discovery) | ✅ Compiles |
+| `uet_oracle` | 5 source files (verifiers, registry) | ✅ Compiles |
+
+Fixed: Cargo.toml dependency paths (`../uet_chain` → `../../uet_chain`)
+
+### OmegaSearch Test Results
+
+```
+Query: "galaxy rotation curve dark matter"
+Result 1: Cluster Virial Data (Ω-dist: 0.0000, axiom overlap: 100%)
+Result 2: Chandra A133 (Ω-dist: 0.0000, axiom overlap: 100%)
+Result 3: Pioneer Anomaly Data (Ω-dist: 0.0000, scale match: ✅)
+```
+
+### Complete task list for today
+
+1. ✅ Git cleanup, Docker/Prisma, CI/CD, repo hygiene
+2. ✅ Agent infra (CORS, Dockerfile, compose, env vars)
+3. ✅ API key security (.env, config.toml untracked)
+4. ✅ LLM integration verified end-to-end
+5. ✅ Knowledge ingestion (116 docs → 5984 chunks)
+6. ✅ LLM model switch (GLM 4.7 Flash)
+7. ✅ LanceDB vector schema fix (fixed_size_list)
+8. ✅ OmegaSearch verified working
+9. ✅ 4 Rust modules added to workspace and compiling
+10. ✅ Dependency paths fixed for all modules