The problem
Ahmed just quit.
3 years on the team. He built the payment system, the retry logic,
the entire auth flow. Next month he's in Toronto.
You have one week to extract everything from his head.
87%
of every AI coding session goes to explaining existing code —
not writing new code. You're paying to re-explain the same
codebase to every developer, on every session, indefinitely.
Why nothing works
Every solution you've tried fails for the same reason.
Write better docs
Nobody does. Out of date by the next sprint.
Shadow the senior dev
Doesn't scale. Ahmed has two weeks left.
Paste into ChatGPT
You pay for 500 pages per question. It forgets tomorrow. Same cost again.
Hire more seniors
200,000 skilled workers left Pakistan in 2024. Good luck retaining them.
Knowledge
shouldn't live
in a person.
It should live
in the codebase.
The product
DevLens
Ingest your codebase once. Answer every developer's question — forever.
Developer asks
"When a download fails halfway, how does the retry logic decide what to do?"
DevLens answers
The retry decision lives in YoutubeDL.urlopen(). It checks
the attempt count against retries, inspects the HTTP status code,
and delegates to FileDownloader.retry() for exponential backoff.
↗ downloader/common.py · line 412
The numbers
90% reduction in cost per question.
Business model
DevLens is a deliverable, not just a tool.
When you hand a client their product, you also hand them a queryable knowledge base of the entire codebase. Permanent. Local. Theirs.
1
Per-project delivery fee — premium line item bundled into the contract
2
Annual seat licence — client's team queries their own codebase
3
Maintenance retainer — DevLens updated as the codebase evolves
Who needs this now
Karachi's software houses have the oldest codebases and the highest turnover.
TPS
80% of Pakistan's banking infrastructure. Code is decades old.
NetSol Technologies
NASDAQ-listed since 1999. Leasing software running 20+ years.
Systems Limited
Sells legacy modernization as a core service. 600+ projects.
Techlogix
Banking and healthcare — long-lived codebases, strict compliance.
i2c Inc.
Payment processing. Old, complex, mission-critical systems.
Where we are
Working prototype. Real results.
✓Full pipeline on yt-dlp — 15,000+ lines of Python, live and queryable today
✓AST chunking outperforms fixed-size and line-break baselines on retrieval accuracy
✓100% local — zero client code leaves the machine. NDA-safe by default.
✓Docker deliverable — drops into any client handoff, no setup needed
✓Bahria University Final Year Project — formally evaluated, peer-reviewed
The team
We felt this problem before we built the solution.
Moiz Ali
Final Year, Computer Science
Bahria University, Karachi
We're not here to sell anything today.
We're here because we think you've felt this problem too —
and we'd like 20 minutes to show you what we built.
View Prototype →