Hey !! I am actively looking for a remote full time engineer role centered around Backend, Infra or Full Stack.
I have been building and doing backend related paid gigs, projects and also I love to write about terminologies and system around backend and infrastructure. I am very much equipped with Golang and many different tech stacks like PostgreSQL, Redis, Nginx which helps build for the scale and large systems
Few Projects to consider
>> OpenTelemetry Bridge Sidecar : Its transparent reverse proxy that injects distributed tracing into HTTP applications without having some major code changes. Backpressure layer for tcp/upd streams : it adjusts the data flow rate between services based on both network conditions something like RTT and packet loss
Implemented 2 phase commit paper purely in golang
>> Wrote a DAG benchmarking tool something like Netflix/P2pLabs repository wrote different IPLDs and benchmarked their performances
>> Built Analytics Engine using Go, Graphana, Redis and docker
>> Apizy, a visual API builder . One can connect build api's in raw nodeJs and expressJs by just dragging and dropping the code blocks
There are many projects like this which was built with the motive to learn something new and to push me from my comfortable zone. If your team is looking for someone who is building in the same space and you think we are aligning I’d love to connect and talk more about my past work. My DMs are open
here are my answers for this for what i understood
1. I would like to decompose some steps into verifiable checkpoints like for a coding agent this might be (i) files read correctly correctly (ii) relevant functions identified (iii) edit applied (iv) tests pass and give a score for each milestone which should be not dependent on each other...
But imho extremely long horizon tasks with extreme accuracy is still an unsolved problem (opinions will change with new facts)
2. In both the interviews I was not able to answer this questions so this is mostly answered by the engineers who are actually working with agents in prod (YESS MY SKILL ISSUE)
my answer : [[ planner (which decides to what to do next) ]] ===> [[ executor (tool call) ]] ==> [[ reflector (did it work??) ]] and then back to first step if not done/not stuck
soo basically I got two complete different answers for the same questions
> (i) using docker + cloud service
(ii) for data ... in healthcare go for HIPAA and in finance go for PCI DSS and FINRA
(iii) separate business logic from the low level working to build a prod grade agent ... something like [[https://t.co/D7IMM9lHi4]]
>> in my second int.. I came with this answer and the engineer said "it won't scale" when he asked this in return I asked a simple question "when you say scale at what point you believe it will break ??"
(i) Hard failures on tools call now you must build a debug loop for the initial loop. What if you want to coordinate among agents in a scatter and gather to improve performance and throughput? You are stuck in the loop. How and when do you break for human in the loop and short circuit work and rollback to a certain point? This approach works for demos for production you have to think about ways things will break and disappoint users
thats all ...
I read about agents from Building Effective Agents blog by @AnthropicAI on agents
idk about others but the last two interviews I gave ... the questions were simple related to ai + infra
> my opinions on long horizon eval like how will i evaluate a agent that is running 20+ steps to complete a task
> agent architecture on prod level ... like a prod ready agent architecture and how do I stop a infinite loops
Thanks for supporting my blogs page it went numbers whenever I dropped them :)
Writing and sharing it publicly really gave me variety of offers ... from SDE to marketer to Tech Lead at some mid level orgs and lastly technical writer ... Really appreciate those times
Last few months went in writing more cooperate blogs for internal toolings and also got teaching gigs through this blog page ... I will be more active in my blog's page from now on as I am not taking anymore cooperate writings and teaching gigs (not really my game)
in few days I will be dropping serverless blog few outlining is only left for it to complete .... Hope y'all like it :)
really want to write a long post on it but I can't find much time so here is a short summary of what I understood and said
1. >I would like to decompose some steps into verifiable checkpoints like for a coding agent this might be (i) files read correctly correctly (ii) relevant functions identified (iii) edit applied (iv) tests pass and give a score for each milestone which should be not dependent on each other...
2. > you can checkout this blog by anthropic on building agents
https://t.co/6Qqt5nL0vR
wait man I am now drafting a really big post for this
idk about others but the last two interviews I gave ... the questions were simple related to ai + infra
> my opinions on long horizon eval like how will i evaluate a agent that is running 20+ steps to complete a task
> agent architecture on prod level ... like a prod ready agent architecture and how do I stop a infinite loops
really want to write a long post on it but I can't find much time so here is a short summary of what I understood and said
1. >I would like to decompose some steps into verifiable checkpoints like for a coding agent this might be (i) files read correctly correctly (ii) relevant functions identified (iii) edit applied (iv) tests pass and give a score for each milestone which should be not dependent on each other...
2. > you can checkout this blog by anthropic on building agents
https://t.co/6Qqt5nL0vR
wait man I am now drafting a really big post for this
really want to write a long post on it but I can't find much time so here is a short summary of what I understood and said
1. >I would like to decompose some steps into verifiable checkpoints like for a coding agent this might be (i) files read correctly correctly (ii) relevant functions identified (iii) edit applied (iv) tests pass and give a score for each milestone which should be not dependent on each other...
2. > you can checkout this blog by anthropic on building agents
https://t.co/6Qqt5nL0vR
wait man I am now drafting a really big post for this