Rishabh Poddar (@rishpoddar) 's Twitter Profile
Rishabh Poddar

@rishpoddar

Engineering at Miko.ai | Founder and Ex-CTO at @supertokensio (open source auth) | Y Combinator S20

ID: 1107932534980014082

calendar_today19-03-2019 09:11:15

485 Tweet

363 Takipçi

15 Takip Edilen

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 12 of building autolead.trythis.app Today I implemented post fetching using the Snoowrap SDK. For each subreddit, I fetch up to 100 posts from the “new” section, stopping once I hit a post older than what I previously scraped. This runs every 10 minutes. To avoid

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 13 of building autolead.trythis.app Today I didn't code, but instead, I researched how to match subreddit posts with a user’s website content. The pipeline I’m considering: Use a cheap (or free) vector-based semantic match. Embed both the Reddit post and the website

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Taking a break from coding now and then is as important as making progress everyday! Your mind needs time to pause, reflect, think, play, and rest.

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 15 of building autolead.trythis.app Started building the post matching pipeline! Step 1: fetch Reddit posts that match the topics of a user’s website. - Both website + post embeddings are already in Postgres - Run a SELECT query (per subreddit) ordering by cosine

Day 15 of building autolead.trythis.app

Started building the post matching pipeline!  

Step 1: fetch Reddit posts that match the topics of a user’s website.  
- Both website + post embeddings are already in Postgres  
- Run a SELECT query (per subreddit) ordering by cosine
Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Just tried out Anthropic Claude code today. Using that for making larger changes + cursor for tab completion or small changes is IMO the optimal vibe coding experience.

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Finally got QLORA fine tuning of Llama 8B working!! Wohooo! 🙌🙌. Will open source my script and pipeline fully. It’s a project to fine tune the LLM to reply as me on my email threads.

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 16 of building autolead.trythis.app Did competitor analysis today: - Signed up for 20 alternative websites - Went through their flows, noting issues and good ideas - This helped me rethink some of my planned flows and clarified how Reddit integrations can work

Day 16 of building  autolead.trythis.app  

Did competitor analysis today:  
- Signed up for 20 alternative websites  
- Went through their flows, noting issues and good ideas  
- This helped me rethink some of my planned flows and clarified how Reddit integrations can work
Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 17 of building autolead.trythis.app Today I integrated Google Gemini App and langfuse.com observability into the stack. Integrating Langfuse with Gemini wasn’t straightforward (unlike their integration with OpenAI SDK), but after a lot of Googling, I got it working (code

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 18 of building autolead.trythis.app Didn't have much time today but made a small improvement: added a DB cache for LLM prompts to save on AI costs. Flow: - DB table: prompt (varchar(100)), model, response - Store key as {prompt.slice(0, 10)}{sha256(prompt)} so

Day 18 of building autolead.trythis.app  

Didn't have much time today but made a small improvement: added a DB cache for LLM prompts to save on AI costs.  

Flow:  
- DB table: prompt (varchar(100)), model, response  
- Store key as {prompt.slice(0, 10)}{sha256(prompt)} so
Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 19 of building autolead.trythis.app Today I optimized a GPT prompt. This prompt takes a Reddit post, the user’s website summary, the matching score, and user feedback, then generates additional criteria to improve the main matching prompt. Process: - Wrote the

Day 19 of building autolead.trythis.app  

Today I optimized a GPT prompt.  

This prompt takes a Reddit post, the user’s website summary, the matching score, and user feedback, then generates additional criteria to improve the main matching prompt.  

Process:  
- Wrote the
Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 20 of building autolead.trythis.app Today I added a new onboarding step. When a user enters their website URL, the system generates a summary of their product and shows it to them. They can edit it if needed. Once confirmed, I use the summary to find relevant

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Spending at least 30 mins walking everyday is one of the best life hacks I know. It refreshes your mind, keeps you healthy, gives you time to reflect, you can spend it hearing a podcast, and you feel so energised after!

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Just finished reading the book: man’s search for meaning. A few points that struck me the most: - the meaning of life is personal and can be found in wanting to achieve a goal, loving someone, of suffering. - humans are capable of enduring almost any suffering as long as they

Rishabh Poddar (@rishpoddar) 's Twitter Profile Photo

Day 21 of building AutoLead Added a new piece to the Reddit post matching pipeline. The pipeline now takes a user_id: - Fetches all subreddits the user is watching - For each subreddit: - Finds recent posts - Runs the matching logic on each post - Saves the

Day 21 of building AutoLead

Added a new piece to the Reddit post matching pipeline.  

The pipeline now takes a user_id:  
- Fetches all subreddits the user is watching  
- For each subreddit:  
  - Finds recent posts  
  - Runs the matching logic on each post  
  - Saves the