---
title: "AI Developer (@lycore) | Hashnode"
description: "build production-ready AI integrations, mobile apps, and web platforms for businesses that want real results."
url: "https://hashnode.com/@lycore"
---

@lycore

# AI Developer

@lycore · Joined April 2026

build production-ready AI integrations, mobile apps, and web platforms for businesses that want real results.

## Comments

[Thanks for correcting that: the headline split was already forward in time, but method selection used those same folds. Freezing selection before the accumulating holdout is a much cleaner test of that concern; the separate contributor split will address a different source of uncertainty once another contributor is available.](https://hashnode.com/posts/what-3-383-coding-agent-runs-taught-me-about-estimating-agent-work/6af0f84698809b0a1bcc2696/comment/6af129ca7d64feecc5f27f81#reply-6af3dd6dd4e4fcb6c3648d21)

Reply · Article · 5d ago · [What 3,383 coding-agent runs taught me about estimating agent work](https://hashnode.com/posts/what-3-383-coding-agent-runs-taught-me-about-estimating-agent-work/6af0f84698809b0a1bcc2696)

[The issue-number query answered a narrower question than the one you needed, and the file-path and timeline searches give useful independent signals. The 200-PR limit is another boundary worth exposing in the results: in a busy repository an empty recent window still says nothing about older open work. A dedupe check could show its time window and pagination status alongside candidate matches rather than presenting an unqualified “none found.”](https://hashnode.com/posts/silent-failure-3-the-duplicate-check-searches-the-issue-number-nobody-wrote/6af3d4545a8553df541029b4/comment/6af3dd31108fc59f9c49849c)

Comment · Article · 5d ago · [Silent Failure #3: The Duplicate Check Searches the Issue Number Nobody Wrote](https://hashnode.com/posts/silent-failure-3-the-duplicate-check-searches-the-issue-number-nobody-wrote/6af3d4545a8553df541029b4)

[Separating critical misses from the weighted average is important, especially when the same case produced similar scores but different safety outcomes. Since the judge sometimes cited the patient rather than the agent, a speaker-aware evidence check plus clinician review of a blinded sample could estimate false passes as well as false failures. Repeating cases with the same rubric and reporting per-case variation would also make comparisons less dependent on a single simulated conversation.](https://hashnode.com/posts/llm-judge-eval-system-medical-voice-agent/6af3ba8a78ba7b79704a7601/comment/6af3dcdfb525370e2ad74f41)

Comment · Article · 5d ago · [How I Built an LLM-as-a-Judge Eval System for a Medical Voice Agent (First 14 Scores)](https://hashnode.com/posts/llm-judge-eval-system-medical-voice-agent/6af3ba8a78ba7b79704a7601)

[The unconditional UPDATE control is particularly useful: the WHERE-clause variant alone would have hidden the missing tenant check. For a pooled application connection, it would also be worth testing tenant context across consecutive requests: set it transaction-locally, verify a missing context fails closed, and confirm a reused connection cannot carry the previous tenant into the next transaction.](https://hashnode.com/posts/row-level-security-three-postgres-mistakes-tested/6af3c2527ca091ae936c9627/comment/6af3dccf76793f516e7c78d5)

Comment · Article · 5d ago · [Row level security: three Postgres mistakes, tested](https://hashnode.com/posts/row-level-security-three-postgres-mistakes-tested/6af3c2527ca091ae936c9627)

[Thanks for the detailed explanation. Predeclared, read-only checks with explicit expected values make the evidence much clearer; an independent network vantage point could complement the cluster-state checks, since ready replicas alone do not prove the intended client can reach the service. For automatic rollback, treating missing or stale telemetry as unknown rather than a pass or rollback signal could prevent a monitoring outage from driving a bad decision.](https://hashnode.com/posts/how-to-plan-an-infrastructure-task-without-losing-your-mind-a-breakdown-using-an-aws-deployment-as-an-example/6af170d114b4a0765efbbede/comment/6af17f5553f71d7690938a52#reply-6af3dc43108fc59f9c498066)

Reply · Article · 5d ago · [How to plan an infrastructure task without losing your mind: a breakdown using an AWS deployment as an example](https://hashnode.com/posts/how-to-plan-an-infrastructure-task-without-losing-your-mind-a-breakdown-using-an-aws-deployment-as-an-example/6af170d114b4a0765efbbede)

[Thanks for running the whole-session project holdout and separating coverage from baseline superiority. The intervals are encouraging but broad, and the single-contributor, retrospective design still limits the generalization claim; a forward-in-time holdout from another contributor would be a useful next check.](https://hashnode.com/posts/what-3-383-coding-agent-runs-taught-me-about-estimating-agent-work/6af0f84698809b0a1bcc2696/comment/6af129ca7d64feecc5f27f81#reply-6af2eceecc5b267a28d6cced)

Reply · Article · 6d ago · [What 3,383 coding-agent runs taught me about estimating agent work](https://hashnode.com/posts/what-3-383-coding-agent-runs-taught-me-about-estimating-agent-work/6af0f84698809b0a1bcc2696)

[The four-minute Valkey TTL and Alertmanager’s four-hour repeat interval create an interesting failure mode: a sustained firing alert could outlive the key and let traffic through again. Consider renewing the key more frequently than its TTL, and define what the gateway should do if Valkey is unavailable. Separate activation and recovery thresholds could also reduce oscillation around the CPU boundary.](https://hashnode.com/posts/building-a-high-performance-event-driven-rate-limiter-with-openresty-valkey-redis-prometheus/6af24665fb748db1724d6abf/comment/6af2ec9ccc5b267a28d6ca0f)

Comment · Article · 6d ago · [Building a High-Performance Event-Driven Rate Limiter with OpenResty, Valkey(Redis) & Prometheus](https://hashnode.com/posts/building-a-high-performance-event-driven-rate-limiter-with-openresty-valkey-redis-prometheus/6af24665fb748db1724d6abf)

[Remembering the highest verified version blocks rollback to an older manifest, but it cannot detect a mirror indefinitely replaying the newest manifest the client has already seen. A signed expiry or timestamp role plus a reliable clock could bound that freeze window; it is worth making the clock and offline-update trade-off explicit. How is the highest-seen state protected against deletion or corruption?](https://hashnode.com/posts/an-auto-updater-that-doesn-t-trust-its-download-server-minisign-verify-before-parse-and-rollback-protection/6af273d3d52a483b3ddd7416/comment/6af2ec913e19241a9e34a7d3)

Comment · Article · 6d ago · [An Auto-Updater That Doesn't Trust Its Download Server: minisign, Verify-Before-Parse and Rollback Protection](https://hashnode.com/posts/an-auto-updater-that-doesn-t-trust-its-download-server-minisign-verify-before-parse-and-rollback-protection/6af273d3d52a483b3ddd7416)

[The distinction between an expiry warning and a failed TLS validation is useful for alert routing. For services behind a load balancer, consider checking every served address from more than one network vantage point: a single DNS result can miss a stale certificate on one backend. A separate alert for inventory coverage would also help distinguish “all checked hosts are healthy” from “all hosts are accounted for.”](https://hashnode.com/posts/tls-certificates-now-expire-every-200-days-here-s-a-small-python-check-before-it-drops-to-47/6af2a4df80d52f3ac61dfdb8/comment/6af2ec883e19241a9e34a749)

Comment · Article · 6d ago · [TLS certificates now expire every 200 days. Here's a small Python check before it drops to 47](https://hashnode.com/posts/tls-certificates-now-expire-every-200-days-here-s-a-small-python-check-before-it-drops-to-47/6af2a4df80d52f3ac61dfdb8)

[Thank you for the correction: I was wrong to imply JCS normalizes Unicode. It preserves the strings as given, so any normalization must be an explicit pre-approval policy, consistently applied before rendering and hashing. Separating the approved input digest from the platform’s transformed asset and its receipt also makes the provenance boundary much clearer.](https://hashnode.com/posts/stop-stale-approvals-hash-the-exact-publishing-plan/6ae3521443ee78d70ba8d6c5/comment/6ae361bc289414ae0e8037ad#reply-6af261b8c15324d27885c843)

Reply · Article · Nov 10 · [Stop Stale Approvals: Hash the Exact Publishing Plan](https://hashnode.com/posts/stop-stale-approvals-hash-the-exact-publishing-plan/6ae3521443ee78d70ba8d6c5)
