rony-chat-bot/internal/portfolio/chunker_test.go
Victor Hugo Vargas 129809067b feat(rag): hybrid retrieval, reference documents, and vendor sampling
Answers were short, sometimes in the wrong language, and occasionally about
projects that do not exist. Measured on a 20-question battery against the real
corpus in both Spanish and English, this takes grounded content from 3/10 to
9/10 and language matching from 7/10 to 10/10.

Retrieval
- Fuse FTS5 keyword search with dense vectors via Reciprocal Rank Fusion.
  Both halves are load-bearing: the corpus is English and visitors ask in
  Spanish, so the meaningful words score zero. "paga" appears 0 times in a
  document that says "Payments: Stripe" — the question "¿Con qué se paga en la
  tienda de ropa?" retrieved nothing at all. Embeddings put all three of that
  project's chunks on top. RRF ranks by agreement rather than comparing a BM25
  score against a cosine, quantities with no shared scale.
- internal/embed: OpenAI-compatible embeddings client, unit-normalised so a
  dot product is the cosine. Reorders by the response `index` field.
- Store a content hash beside each vector and skip rows where it no longer
  matches the chunk. Chunk ids survive body edits, so without this an edited
  document keeps serving embeddings that describe text that is gone —
  reproduced live by changing a payment provider and watching the old one keep
  coming back.
- Degrade to keyword-only when the embedder is down instead of failing.

Reference documents that are not projects
- Index `.mdx` alongside `.md`, and split sources into projects (announced in
  the catalogue) and reference material (retrievable, never listed). A CV is
  what someone deciding whether to hire actually reads, and it was unreachable
  while it lived only in the Astro site — but filing it under projects made
  the bot list "cv" as one of Victor's works.
- Skip each directory's README. `data/projects/README.md` was being indexed,
  so the catalogue injected into every prompt announced "README" and
  "README.es" as projects of Victor's.
- Exclude frontmatter from retrieval. It is dense metadata in a very short
  chunk, which makes it a magnet for short queries: a CV's `location:` field
  answered "¿Dónde ha trabajado Victor?" with a city instead of a work history.
- Split oversized sections at `###` before falling back to byte offsets. A CV's
  Experience section is a list of jobs, and size-splitting cut one mid-word,
  stranding the employer's name in the previous chunk.

Prompt and sampling
- Inject the full project catalogue every turn. Top-K search returns the best
  matching sections, so "list every project" cannot be answered from retrieval
  alone, and a small model asked to enumerate from partial hits invents the
  rest. ~10 tokens per project; this is what stopped the invented names.
- Wire the sampling parameters the model authors publish (top_k, top_p, min_p,
  repeat_penalty, presence_penalty) through config to llama.cpp. Leaving them
  at llama.cpp's defaults produced 16-token stub answers.
- Localised system prompt selected by detected language. The English prompt
  plus "reply in the user's language" answered 1/5 Spanish questions in
  Spanish; few-shot examples fixed the language but got copied verbatim into
  real answers.
- Fold compaction's system notes into the leading system message. Gemma's chat
  template rejects a system message that is not first, and the whole request
  failed with HTTP 400 the moment compaction fired.

Configuration and docs
- context_size 4096, down from 8192. The largest prompt this bot ever built
  over 20 real requests was 1255 tokens, compaction starts at ~3070, and the
  cut saved 212 MB resident with zero truncations and identical throughput.
- Correct the RAM figures throughout. They were measured with a GPU absorbing
  llama.cpp's buffers; on a GPU-less VPS those come out of system RAM, which
  is 1.1 GB more for qwen2.5-3b and 2.8 GB more for granite. Both READMEs
  still started gemma-3-1b while the config defaulted to qwen, and neither
  started the embedder at all.

Measured on the 2-core, 8 GB CPU-only target: 3.64 GB LLM + 0.91 GB embedder
+ 0.02 GB bot, 21.0 tok/s steady state.

Known and unfixed, so they are not re-filed as new bugs: the model reads dates
out of the CV correctly but does the arithmetic on them wrong, and "¿Dónde ha
trabajado Victor?" still answers with projects rather than employers, though
"¿En qué empresas ha trabajado?" works.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-07-30 15:45:36 -07:00

155 lines
3.8 KiB
Go
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

package portfolio
import (
"strings"
"testing"
)
func TestSplitMarkdownSections(t *testing.T) {
cases := []struct {
name string
input string
wantCount int
wantHeading []string
}{
{
name: "frontmatter only",
input: `---
title: "Foo"
tags: ["go"]
---
# Foo
Body of foo.`,
wantCount: 2,
wantHeading: []string{"frontmatter", "Foo"},
},
{
name: "no frontmatter, multiple H2s",
input: `# Title
intro paragraph
## Description
description body
## Tech stack
- Go
- SQLite`,
wantCount: 3,
wantHeading: []string{"Title", "Description", "Tech stack"},
},
{
name: "H3 stays under parent H2",
input: `# T
## Section
content
### H3 detail
H3 content stays here`,
wantCount: 1, // T has no body → dropped; H3 content folds into Section
wantHeading: []string{"Section"},
},
{
name: "drop empty section",
input: `## X
ok
## Empty
## Y
content`,
wantCount: 2,
wantHeading: []string{"X", "Y"},
},
{
name: "long H2 sub-splits",
input: "## Long\n" + strings.Repeat("a ", 800),
wantCount: 2, // 2 sub-splits of Long
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
got := SplitMarkdownSections(c.input, DefaultChunkerConfig())
if len(got) != c.wantCount {
t.Errorf("got %d chunks, want %d. Headings: %v", len(got), c.wantCount, headings(got))
}
if c.wantHeading != nil {
if !equalSlice(headings(got), c.wantHeading) {
t.Errorf("headings = %v, want %v", headings(got), c.wantHeading)
}
}
})
}
}
func headings(sections []section) []string {
out := make([]string, len(sections))
for i, s := range sections {
out[i] = s.Heading
}
return out
}
func equalSlice(a, b []string) bool {
if len(a) != len(b) {
return false
}
for i := range a {
if a[i] != b[i] {
return false
}
}
return true
}
// An oversized section made of H3 entries (a CV's Experience list, a FAQ)
// must split per entry, not per byte. Size-splitting used to cut a job entry
// mid-word and strand the employer name in the previous chunk.
func TestSubSplitPrefersH3BoundariesOverByteOffsets(t *testing.T) {
body := "### Metrimex — Frontend Developer\n\n#### Jul 2024 Jun 2026\n\n" +
strings.Repeat("Built an access-control app for physical sites. ", 12) +
"\n\n### Didcom — Android Developer\n\n#### Jul 2023 Jul 2024\n\n" +
strings.Repeat("Built an Android app for an on-demand ride-sharing service. ", 12) +
"\n\n### Teknol — Software Developer\n\n#### Jun 2016 Jan 2017\n\n" +
strings.Repeat("Maintained an internal tool. ", 12)
got := subSplit([]section{{Heading: "Experience", Body: body}}, 400, 40)
if len(got) != 3 {
t.Fatalf("got %d chunks, want one per job:\n%+v", len(got), got)
}
for i, want := range []string{"Metrimex", "Didcom", "Teknol"} {
if !strings.Contains(got[i].Heading, want) {
t.Errorf("chunk %d heading = %q, want it to name %s", i, got[i].Heading, want)
}
if !strings.HasPrefix(got[i].Body, "### "+want) {
t.Errorf("chunk %d body should start at its own heading, got %.40q", i, got[i].Body)
}
}
// No chunk may begin mid-sentence — that is the defect this replaces.
for i, s := range got {
if strings.HasPrefix(s.Body, "... ") {
t.Errorf("chunk %d still starts with a byte-split continuation: %.40q", i, s.Body)
}
}
}
// A long section with no H3 structure still falls back to size splitting.
func TestSubSplitFallsBackToSizeWithoutH3(t *testing.T) {
body := strings.Repeat("plain prose with no subheadings at all. ", 40)
got := subSplit([]section{{Heading: "Description", Body: body}}, 300, 30)
if len(got) < 2 {
t.Fatalf("expected the oversized section to be split, got %d", len(got))
}
for _, s := range got {
if s.Heading != "Description" {
t.Errorf("size-split chunk changed heading to %q", s.Heading)
}
}
}