Samyak Gupta

Building AI in public
100 Days of AI | GenAI • Agents • Backend
Sharing what I learn & build.

[Samyak Gupta @SamyakLabs]

Let me introduce myself — August 2026

CS graduate trying to figure things out, build things that matter, and get better every day. Currently diving deep into AI/GenAI, backend development & cybersecurity while sharpening my DSA and interview skills.

[Samyak Gupta @SamyakLabs]

Day 17/100
Cleaned up the research agent UI and started a new project.
Improved the frontend of the research agent so it feels less rough.

[Samyak Gupta @SamyakLabs]

Also started something new: Give it a public GitHub repo URL -> it clones the code, runs security scanners, investigates findings with an LLM, filters false positives, and returns an evidence-backed security review.
Still early, but the idea feels solid.

#AI #Agents #Security

[Samyak Gupta @SamyakLabs]

Day 17/100
Cleaned up the research agent UI and started a new project.
Improved the frontend of the research agent so it feels less rough.

[Samyak Gupta @SamyakLabs]

Day 16/100
Finally finished the research agent I was building from scratch.

[Samyak Gupta @SamyakLabs]

The UI is still pretty basic (not great), but the core system actually works end to end.

Feels good to close this one out. On to the next thing.

#AI #Agents #Python

[Samyak Gupta @SamyakLabs]

Day 16/100
Finally finished the research agent I was building from scratch.

[Samyak Gupta @SamyakLabs]

Wrapped it up today
Added conversation memory, wrote tests and evals, and got the whole thing running in Docker.
It can plan research, search the web, read sources, extract evidence, keep track of what it’s done, and give a final answer - all without LangChain or LangGraph.

[Samyak Gupta @SamyakLabs]

@MBashirzadeh got it will test this

[Samyak Gupta @SamyakLabs]

Day 15/100
Started adding real memory and context management to the research agent.

[Samyak Gupta @SamyakLabs]

Also spent some time exploring Hermes agent and a few other agent systems to see how they handle memory, tool use, and long-running workflows.

Still early on the memory side, but it’s becoming clear how much difference proper context management makes.

#AI #Agents #Python

[Samyak Gupta @SamyakLabs]

Continued working on the project today mainly focused on conversation memory and better context handling so the agent doesn’t lose track across steps.

[Samyak Gupta @SamyakLabs]

- Added completion tracking so the agent knows what’s done and what’s left

Removed the need for the main agent to manually save findings.

Still debugging a few things, but the structure feels much cleaner now.

#AI #Agents #Python

[Samyak Gupta @SamyakLabs]

Day 14/100. Continued building the research agent — focused on structured state and automatic evidence extraction.

[Samyak Gupta @SamyakLabs]

- Separated messages (execution history) from structured research memory
Also added automatic evidence extraction:
- After reading a source, a separate Groq call pulls out only the useful evidence
- Findings get stored in state automatically

[Samyak Gupta @SamyakLabs]

@afqxq79143289 That’s awesome. Just went through the site — looks like the pipeline is fully automated.
Curious how you handle quality control. Do you still have any human review at certain stages, or is it completely hands-off?

[Samyak Gupta @SamyakLabs]

Day 13/100
Put the courses aside and started building a research agent from scratch.

[Samyak Gupta @SamyakLabs]

Still a work in progress. Will take a few more days to get it in good shape.

Main realization today: courses only take you so far. At some point you just have to sit down and actually build the thing.

#AI #Agents #Python

[Samyak Gupta @SamyakLabs]

Building a tool-calling research agent in pure Python that can:
- Dynamically select web-search and data-processing tools
- Maintain conversation state
- Validate outputs with Pydantic
- Expose the agent through FastAPI