We built an autonomous multimodal agent ecosystem for technical evaluation.
1. Monitoring a candidate.
2. Support for multiple languages with an adaptive approach.
3. Live interview with an agent.
4. Live conversation with an assistant.
#GeminiLiveAgentChallenge
Built LivePoly, a Monopoly-inspired board game.
Server-authoritative game engine keeps players synced via https://t.co/TomGEHFwno. PostgreSQL handles persistence, while Redis + BullMQ power bots and turn timers.
Running on AWS with a React PWA frontend.
https://t.co/RdzLODkfAG
@TechWimps@SumitM_X there are scenarios where kafka is probably an overkill lol. for simple fanout notifs where we don't care about if the listeners are active, just use redis pub sub.
for verification notifs after processing a provider webhooks, use pub sub + sticky keys.
I'll make more money but I'll never be this age again. I don't want to look back at my twenties and not have any memories of myself having fun and living it. Don't want to spend it all in front of a computer.
Added Rate limiting to my Endpoints using redis.
Redis INCR increments the counter, the first request EXPIRE sets 60 second window. If counter > limit return 429 Too Many Requests
Limits per route type:
→ /auth/login → 5 requests/min (brute force prevention)
→ /auth/forgot → 3 requests/min
→ /menu → 100 requests/min (public browsing)
If Redis is down, request passes through. its better to serve requests than block everyone
@abolorreeeee I send 30 requests at 5:59:59 PM and another 30 at 6:00:00 PM. Your limiter sees two valid windows and allows all 60 requests, even though they arrived within a second or so.
Fixed windows also don't handle bursts well since I can spam requests while staying within the limit.
@abolorreeeee Fixed window rate-limiters has a boundary problem. It resets based on time intervals while sliding window tracks requests over the set time continuously.
Now imagine you have an endpoint with a fixed window limit of 30 requests per minute. I can simply exploit this easily ->
Won Google's Gemma Hackathon on DevTo alongside my friends @a_simie and another(He's not active on Twitter)
We built a fully autonomous Android accessibility agent powered by the Gemma 4 2B open-weight model.
1/main to go.
My approach would be to aggregate notifs using a time window + rate threshold by storing post interactions in Redis. Low rate? send notifs 1 by 1. If the rate crosses the threshold, batch the notifs.
Also, notifs would go through an event queue and be processed by async workers.
System design challenge.
On Twitter, if your post gets 10 likes in 30 minutes for example, they send you the notifications one after the other but when your post goes viral and you're getting tens or hundreds of likes every minute, they start grouping it, "{user} and 100 others liked your post".
How do you implement that kind of notification system?
Currently building LivePoly: my own online, Monopoly-inspired game experience.
Already designed the system architecture and built the core game engine, rooms, realtime gameplay, bots, turn timers, game recovery, snapshots and results.
A lot left to do.
https://t.co/uFTSWAIgAo