Slides: We Cut 94% of AI Coding Tokens With a Local Code Index - Rajkumar Sakthivel, Tesco
Source Video
We Cut 94% of AI Coding Tokens With a Local Code Index - Rajkumar Sakthivel, Tesco
Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.
Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides
OCR text:
AIENGINEERWORLD'SFAIR
June30-July2,2026-SanFrancisco
SEARCH&RETRIEVAL TRACK
We Cut 94% of Our Al Coding Tokens
With a Local Code Index
Here's the architecture.
0.4ms
Token Reduction
Search Latency
Recall@10
RajkumarSakthivel
RS
github.com/elara-labs/code-
context-engine
OCR text:
Every Al coding tool we tried had the same assumption:
send as much context as possible.
sw \2)2)
| i“
OCR text:
We optimized the model. We should have optimized the
eet a> 4
@ A retrieval layer between codebase and agent 4»
t
OCR text:
Where your tokens actually go
< | Output compression
XO,
= ~8% off total bill
Input retrieval
= ~61% off total bill
@ 1 .
OCR text:
A local retrieval layer between codebase and agent
Bic ibeCitg Hybrid Poin g Code Confidence
Chunking Retrieval Compression Graph TorLitils]
10 langs 94% a Tica
OCR text:
Why not just vector search?
Ss re a
Vector Search FTS5 (BM25) RRF Fusion
res cree,
Neither retriever is good enough alone. Together they cover each other's blind spots. a
OCR text:
The hardest problem wasn't retrieval. It was knowing when
retrieval was -
Confidence scoring blend
@ Simple heuristic won
OCR text:
BENCHMARK
FastAPI
FastAPl:53files,20realquestions,reproducible
Full file baseline
83,681 tok/q
Afterretrieval
4,927tok/q
retrieval savings
After compression
523tok/q
No cherry-picking.No synthetic queries.
20questionsadeveloperwouldactuallyask
Recall@10
$pythonbenchmarks/run_benchmark.py
--repo fastapi/fastapi--source-dirfa
elara-labs/code-context-engine
OCR text:
TRADE-OFFS
i
What we're honest about
94% is against full-file reads Monorepos dilute recall
Embedding model matters What actually worked
: )
OCR text:
KEY TAKEAWAY
MEASURABLE
Everytokentracked.Everydollarcounted.
my-project.247queries.Last query5mago
90088xtokensaved
Input savings
12.4M
tokens
Output savings
48.2k
tokens
Total saved
12.4M
tokens
Breakdown:
retrieval
10.4H $156.00
chunk conpression
421.5k
48.2k
nates.Actual tokens served vs full
LDo
costs from live modelpricing
Thank you·Rajkumar Sakthivel
OCR text:
The biggest optimization in Al coding
isn't the model. [t's the context.
SA local a
Bese) Try it now
ae
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.