Janady13/missionmatch / internal/modules/ai/retrieval.go

Source identity

Field	Recorded value
Source type	GitHub source
Original record	Janady13/missionmatch / internal/modules/ai/retrieval.go
Date	Not specified in source
Repository	Janady13/missionmatch
Revision	43d56387270ef4da4cabd794d97ca6a49909bc7d
Branch	w3-ai-staffing-recommendations
Path or archive member	internal/modules/ai/retrieval.go
Original SHA-256	54127da76e116f4a7c38483bb90a2644759838840fc82c5f397df19705ec2eac
Git blob SHA	d8533d267ce95fb600abca6a4caf8668dfb95516
Reading-copy SHA-256	54127da76e116f4a7c38483bb90a2644759838840fc82c5f397df19705ec2eac

Original source ↗

Every recorded occurrence

These 17 records contain identical original source bytes. Their paths and revisions remain part of the archive.

Original record	Branch or archive	Revision / identifier	Original location
Janady13/missionmatch / internal/modules/ai/retrieval.go	w3-ai-staffing-recommendations	43d56387270ef4da4cabd794d97ca6a49909bc7d	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ws3-clearance	4b8e7ce594c068621867f9c9deed0ec64ae6c553	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ai-intelligence-layer	5d5ddff70393a197c5f54a45280bb5dd42b24755	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	per-tenant-routing	9bde0f9a74b89d01980fa658e5253acda7528d0c	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ai-brain-live	aac32fc49c878cc5bca099459307611562d0bac1	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ws1-ontology-sync	b596f449a32a1016fbd403c36505cc00a1ac9ee6	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	deploy-and-ops	bdacd9d0b4d5998d25fc74bf4007f4c3ec538c7a	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	phase2-eval-fairness	d361af254701975fe239927656ababb56374655e	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ws2-alias-relationship	e4dada4601c6eb101830f9d223a9655dfa120616	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	per-tenant-routing	9bde0f9a74b89d01980fa658e5253acda7528d0c	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ws1-ontology-sync	b596f449a32a1016fbd403c36505cc00a1ac9ee6	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	w3-ai-staffing-recommendations	43d56387270ef4da4cabd794d97ca6a49909bc7d	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ws3-clearance	4b8e7ce594c068621867f9c9deed0ec64ae6c553	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	phase2-eval-fairness	d361af254701975fe239927656ababb56374655e	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ai-intelligence-layer	5d5ddff70393a197c5f54a45280bb5dd42b24755	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	deploy-and-ops	bdacd9d0b4d5998d25fc74bf4007f4c3ec538c7a	Open original
Janady13/missionmatch / internal/modules/ai/retrieval.go	ai-brain-live	aac32fc49c878cc5bca099459307611562d0bac1	Open original

Recovered source transcript

Download complete source text ↓72 lines · 2.5 KB

package ai

import (
	"context"
	"errors"

	"github.com/jackc/pgx/v5"
	pgvector "github.com/pgvector/pgvector-go"

	"github.com/missionworkforce/missionmatch/internal/platform/database/dbgen"
	"github.com/missionworkforce/missionmatch/internal/platform/inference"
	"github.com/missionworkforce/missionmatch/internal/platform/tenant"
)

// Hit is one retrieved candidate with its fused relevance score.
type Hit struct {
	CandidateID string  `json:"candidate_id"`
	FullName    string  `json:"full_name"`
	Score       float64 `json:"rrf_score"`
}

// HybridSearch is the RAG retrieval leg: it embeds the query through the MEGAMIND plane and runs the
// existing hybrid query that fuses pgvector cosine kNN with full text rank via Reciprocal Rank Fusion
// (docs/AI_RESEARCH.md, pgvector + RRF). It runs RLS scoped as the tenant, so it returns only the
// caller's candidates, and it tunes the HNSW scan per the pgvector 0.8 guidance before the search.
//
// It never touches the eeo schema (the query reads only candidates), and the query string is guarded so
// a demographic reference cannot enter the retrieval path.
func HybridSearch(ctx context.Context, db *tenant.DB, client *inference.Client, query string, limit int) ([]Hit, error) {
	if db == nil || client == nil {
		return nil, errors.New("ai: retrieval needs a database and an inference client")
	}
	if err := ForbidEEO(query); err != nil {
		return nil, err
	}
	if limit <= 0 || limit > 100 {
		limit = 20
	}
	emb, err := client.Embed(ctx, []string{query})
	if err != nil {
		return nil, err
	}
	if len(emb.Embeddings) != 1 {
		return nil, errors.New("ai: query embedding failed")
	}
	qv := pgvector.NewVector(emb.Embeddings[0])

	var hits []Hit
	err = db.RunInTenantTx(ctx, func(ctx context.Context, tx pgx.Tx) error {
		// pgvector 0.8 tuning: a wider ef_search lifts recall on the filtered (RLS) scan, and relaxed
		// iterative scan avoids overfiltering under the tenant predicate (docs/AI_RESEARCH.md, pgvector).
		if _, err := tx.Exec(ctx, "SET LOCAL hnsw.ef_search = 100"); err != nil {
			return err
		}
		if _, err := tx.Exec(ctx, "SET LOCAL hnsw.iterative_scan = 'relaxed_order'"); err != nil {
			return err
		}
		rows, err := dbgen.New(tx).SearchCandidatesHybrid(ctx, dbgen.SearchCandidatesHybridParams{
			Column1: qv, Limit: int32(limit), WebsearchToTsquery: query,
		})
		if err != nil {
			return err
		}
		hits = make([]Hit, len(rows))
		for i, r := range rows {
			hits[i] = Hit{CandidateID: r.ID.String(), FullName: r.FullName, Score: r.RrfScore}
		}
		return nil
	})
	return hits, err
}