Franck SN
@ndzfs
Followers
2,249
Following
1,376
Media
957
Statuses
8,647
Techno-optimist, Engineer, Entrepreneur. Built , Building - open for consulting gigs
Paris
Joined November 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
I VOTE
• 574808 Tweets
#2024MAMAVOTE
• 508886 Tweets
गोपाल मिश्रा
• 125435 Tweets
Libya
• 100257 Tweets
Nobel
• 98975 Tweets
LOVE MONEY FAME OUT NOW
• 70456 Tweets
Artist of the Year
• 70215 Tweets
Super Eagles
• 59351 Tweets
神奈川県警
• 54952 Tweets
アトラス彗星
• 47396 Tweets
Choice - Female Award
• 44799 Tweets
#njkopw
• 44751 Tweets
BOOKEXPO X OHMLENG
• 37161 Tweets
#tomorrowxtogether
• 35083 Tweets
#スポ男
• 32054 Tweets
Daron Acemoğlu
• 27814 Tweets
レモンポップ
• 25062 Tweets
千葉県北西部
• 24519 Tweets
地震速報
• 22804 Tweets
棚橋引退
• 18069 Tweets
棚橋弘至
• 14861 Tweets
ペプチドナイル
• 12369 Tweets
ノミネート
• 10256 Tweets
Simon Johnson
• 10018 Tweets
In Europe, except UK perhaps, you can be good at what you do, people won't take you seriously unless you went to certain schools.
6
138
548
@Carnage4Life
It's a classic move. You hire consultant so they can be held accountable of the decision to layoff. It's a political move, and since they are not internal, they can be more ruthless.
3
10
369
This is probably the best DSPy course on the internet. And I'm frankly not biased at all.
Learn to program, not prompt language models and improve your quality of life in the process 😂
4
52
313
For the most advanced use cases of DSPy, I highly recommend this module. It covers everything from building knowledge graphs to implementing RAG, as well as integrating DSPy with FastAPI to develop user-facing LLM-based applications.
5
37
226
My contribution has been accepted and synthetic data generation has been added to DSPy. 🔥
You can either generate data by defining a pydantic model or by feeding some initial examples. And it is blazingly fast !
Here is how to use it:
I am going even further. Now, with a bit of meta-programming, I can dynamically create signatures on the fly and generate synthetic datasets based on pydantic models. Crazy nice to see it in action 🔥
I propose to add the synthetic_data_generation function in the utils of DSPy.
5
19
155
7
29
193
I built an AI financial analyst using
@LangChainAI
,
@OpenAI
function calling and
@streamlit
(thread 1/)
#buildinpublic
9
30
184
What if I build a new version of
@babyAGI_
using DSPy ?
Let's go !
original github by
@yoheinakajima
:
5
23
185
@PaulLarrouturou
@sudouest
@SO_Paysbasque
C'est pour cela que je n'ai aucune tolérance pour ces pratiques de bizutage à la con.
6
15
174
@Carnage4Life
What I like is that we can now see how they think and realize they aren't that special, have a lot of wrong takes, are terribly wrong on a lot of issues. Kinda eye opening
2
11
164
Holy smoke !! I achieved near perfect coreference resolution using DSPy and my knowledge graph is quite neat now. Sharing the full tutorial on Lycee AI in a few minutes. Link will be under this tweet when I am done. 🔥🔥
How it started vs How It's Going
8
18
165
I am going even further. Now, with a bit of meta-programming, I can dynamically create signatures on the fly and generate synthetic datasets based on pydantic models. Crazy nice to see it in action 🔥
I propose to add the synthetic_data_generation function in the utils of DSPy.
And here it is: a pure DSPy-based implementation of synthetic prompt optimization.
First the synthetic data generation is done using a kind of hacked DSPy signature with a random input field and output fields that match the schema of the data I want to generate. Note the use of
4
16
136
5
19
155
New video just dropped in my module on how to create a full-stack web application using DSPy, FastAPI, Weaviate, and React. In the video, I explain how to set up a basic API using FastAPI and a simple DSPy service. More chapters are coming in the next few days. Link to the full
2
29
149
1
1
144
Why You Should Add A Classification Step In Your Retrieval Augmented Generation System
4
20
146
And here it is: a pure DSPy-based implementation of synthetic prompt optimization.
First the synthetic data generation is done using a kind of hacked DSPy signature with a random input field and output fields that match the schema of the data I want to generate. Note the use of
Now working on a pure DSPy implementation and another one using anonLLM. Stay tuned, we are cooking 🔥
0
1
17
4
16
136
We are moving fast here. The backend is now up and running. Users can upload documents (TXT, PDFs even OCR). The documents are then chunked and stored in our Weaviate vector store. We use a DSPy RAG module to provide answers and the context used to provide these answers.
Here is
2
14
133
Okay, this DSPy + FastAPI tutorial is moving quickly, and it is amazing to see.
I have now implemented a file parser functionality and used the factory method pattern. It is 🔥🔥🔥
Here is a preview.
Full video in the second tweet:
2
13
123
@Carnage4Life
I think at its core it's because people need badly to tell themselves stories about how talented they are. In a world of self promotion, admitting the luck part is a hard pill to swallow for a lot of people. They need to believe in their own mythology.
3
13
110
@jacquincl
Ils partent à l'église tous les dimanches, on leur demande d'aimer leur prochain comme eux mêmes, ils sortent ils veulent chasser les immigrés. Comment fait on pour vivre avec une telle dissonance cognitive ?????
18
11
102
@Tyler_A_Harper
I had the exact same revelation in college. Rich people have to convince themselves that they are simply more intelligent than other people, because it allows them to justify their privilege without feeling bad about it. Truth is they are mostly lucky (inheritance, being born
2
2
101
@edwyplenel
@auroreberge
@AnttonRouget
@Mediapart
et elle est ministre en charge "de la lutte contre les discriminations", sans blagues mdrrr
1
3
96
@Carnage4Life
What an indictment of their business model then, that they can't survive if they have to pay a fair minimum wage. Or am I missing something?
3
3
100
Finally added a React frontend to the RAG application. Still using DSPy, FastAPI, and Weaviate in the backend. It is amazing to develop a full-stack web application powered by DSPy.🔥🔥🔥🔥🔥
Link to module in second tweet !
We are moving fast here. The backend is now up and running. Users can upload documents (TXT, PDFs even OCR). The documents are then chunked and stored in our Weaviate vector store. We use a DSPy RAG module to provide answers and the context used to provide these answers.
Here is
2
14
133
3
10
99
Added two new chapters to the advanced DSPy tutorial 🔥
In the second chapter, I explain how to improve the knowledge graph extractor we built in the first chapter, aiming to reduce duplicate entities.
In the third chapter, I explain what coreference resolution is and why it is
Users from Lycee AI are asking me to create more advanced DSPy tutorials. I'm starting a new series today with an exploration of DSPy and knowledge graphs.
I wrote a script to build a knowledge graph using DSPy. I'll be dropping the video on Lycee AI later today. It's actually
3
4
95
3
15
98
Users from Lycee AI are asking me to create more advanced DSPy tutorials. I'm starting a new series today with an exploration of DSPy and knowledge graphs.
I wrote a script to build a knowledge graph using DSPy. I'll be dropping the video on Lycee AI later today. It's actually
3
4
95
@levelsio
@Kling_ai
@synclabs_so
It's still easy to see that it’s AI-made, but there has been amazing progress on the image and video generation front— even more impressive than what I see with LLMs. Plus, it's quite clear how to monetize it. However, it still requires a lot of commitment because the content
5
2
94
Learn how to build a financial analyst copilot that will help you decide which stocks to buy or sell. Powered by DSPy, OpenAI, and a Weaviate database, this tool enables your agent to remember past information.
What is amazing about this agent is that you can actually access its
2
14
89
I took me 2 minutes to set up the monitoring of my AI financial analysis app with LangSmith (
@langchain
).
Here is how I did it: 1/
#buildinpublic
5
14
85
I've been generating RPG players by the hundreds thanks to DSPy. And all I need is to define a pydantic class like so:
4
5
84
Did a pretty nice implementation of an AI agent using state machines and DSPy.
Will be writing and sharing a full article about it.
7
9
81
From unstructured to structured data using DSPy
Full tutorial on Lycee AI (link in second tweet)
Users from Lycee AI are asking me to create more advanced DSPy tutorials. I'm starting a new series today with an exploration of DSPy and knowledge graphs.
I wrote a script to build a knowledge graph using DSPy. I'll be dropping the video on Lycee AI later today. It's actually
3
4
95
1
10
81
@jessicanono1
Les personnes les plus méchantes et absurdes que j'ai pu rencontrer dans ma vie allaient à l'église tous les dimanches. Les personnes les plus gentilles et généreuses que j'ai pu rencontrer allaient également à l'église tous les dimanches. Je crois que c'est difficile de conclure
0
14
75
LLMs are better than MBA students at strategy 🤯🤯
3
10
72
It's underrated how powerful DSPy can be when you want to create agentic workflows.
And when you add Typed Predictors from
@thomasahle
and
@NormalComputing
in the mix, the results are 🔥
I just built a chess player agent and can watch it play against the stockfish engine. And
7
13
68
The next module on Lycee AI will be lit. 🔥🔥
It will be about developing RAG apps using DSPy, FastAPI, Weaviate and React for the frontend. Who is ready??
Order now at a reduced price on Lycee AI.
I added another chapter to the advanced DSPy module: How to do retrieval augmented generation using DSPy and a
@weaviate_io
vector database.
Here a preview (link to full video in the second tweet):
3
8
44
3
7
66
Chat with your database using natural language.
Learn how to build such a system on Lycee AI (link below)
1
5
64
Just published a beginner-friendly guide on how to use LangSmith to monitor your LLM apps !
Another powerful release by the
@LangChainAI
team !
#LangSmith
#AI
#MachineLearning
#buildinpublic
Check it out here:
1
18
64
Just added a new chapter. Learn how to create custom tools for your AI agents using LangChain.
Link to full video in second tweet
Started building a financial analyst using
@LangChainAI
,
@tavilyai
and
@GroqInc
Link to video in second tweet
3
11
51
1
5
65
Did a basic implementation first, using typed predictors.
Will generalize it later (general def of agents, vector stores, tools etc)
Explanatory article in second tweet.
Link to github in third tweet.
What if I build a new version of
@babyAGI_
using DSPy ?
Let's go !
original github by
@yoheinakajima
:
5
23
185
2
6
58
Another chapter just dropped: adding
@weaviate
into the mix! The RAG module is powered by DSPy, of course. Things are really starting to take shape in the backend. Amazing to see! 🔥🔥🔥
As always, here is the preview (full link to the video in second tweet)
Okay, this DSPy + FastAPI tutorial is moving quickly, and it is amazing to see.
I have now implemented a file parser functionality and used the factory method pattern. It is 🔥🔥🔥
Here is a preview.
Full video in the second tweet:
2
13
123
2
10
55
If you haven't started already, this is your reminded to start learning DSPy on Lycee AI before the AGI singularity
0
10
54
We don't talk enough about spacy-llm. It is so powerful. And when you mix it with DSPy, the results are incredible. I am working on a video that i will add to the "Advanced DSPy" module on Lycee AI. It will be lit 🔥🔥🔥
4
8
54
Started building a financial analyst using
@LangChainAI
,
@tavilyai
and
@GroqInc
Link to video in second tweet
3
11
51
I've created a LLM-based Movie Recommender using
@OpenAI
function calling,
@streamlit
, and
@LangChainAI
. Should you be interested in the Python code that powers this, don't hesitate let me know in the comments !
#buildinpublic
16
7
50
@Rahll
This is what happens when expectations of productivity gains thanks to AI are not realistically set: "Despite 96% of C-suite executives expecting AI to boost productivity, the study reveals that, 77% of employees using AI say it has added to their workload and created challenges
1
0
50
I just finished managing background tasks in my RAG application. Check out a preview here. We are using FastAPI, NLTK, and the OpenAI API directly, without any framework. Link to full video in the second tweet.
1
3
48
I think people are overlooking synthetic data generation and prompt optimization. Thankfully, DSPy is exactly what we need for both. You can unlock many useful applications when data generation is no longer an issue.
1
12
46
Folks, the feature is finally available!
You can now play with the Mistral AI API on DSPy. 🔥
@MistralAI
🫸🫷DSPy
3
8
45
You can now learn how to build BabyAGI by
@yoheinakajima
using DSPy, on !!! Enjoy!!!
BabyAGI is a task-based autonomous AI agent. It captivated the imagination of many AI engineers last year and can be said to be the initial spark to the emergence of
1
8
41
DSPy to the moon 🔥🔥🔥🚀🚀🚀
This is awesome! I am planning to democratize the use of DSPy even more on my LMS platform
I am currently working on a comprehensive tutorial to make DSPy accessible to a lot of AI engineers / developers.
Create an account
3
4
44
I added another chapter to the advanced DSPy module: How to do retrieval augmented generation using DSPy and a
@weaviate_io
vector database.
Here a preview (link to full video in the second tweet):
3
8
44
Amazing to see how active the DSPy community is these days.
@krypticmouse
and I have been working on synthetic data generation in parallel, with two similar but slightly different approaches.
@krypticmouse
's approach: feed some initial data and generate synthetic data based on
Looks like both
@krypticmouse
and
@ndzfs
have been independently developing synthetic dataset generators for DSPy optimizers.
So today we got two PRs for this, with very different approaches! Community feedback welcome; maybe we can merge the two into one robust version?
6
16
111
3
8
42
I have finally completed the tutorial on using Spacy-llm and DSPy for Named Entity Recognition. It concludes with an unexpected twist. Link in the following tweet. Preview:
We don't talk enough about spacy-llm. It is so powerful. And when you mix it with DSPy, the results are incredible. I am working on a video that i will add to the "Advanced DSPy" module on Lycee AI. It will be lit 🔥🔥🔥
4
8
54
4
6
41
@H0wie_Xu
@ycombinator
@sama
@OpenAI
I love Sam, but I see no signs of AGI in the near term. More capable models ? Sure. AGI ? Nope.
6
2
40
@iamharaldur
It's a real problem in Paris indeed. Some efforts are in the works, we will get there hopefully.
0
0
38
Inspired by
@CShorten30
video, I just built a stock analyst using DSPy. More details in my article below !
#buildinpublic
2
7
38
I am in the arena trying stuff using the LangChain Expression Language !
Check it out ! It's so powerful !
@LangChainAI
@chamath
#buildinpublic
2
8
36
How to quickly and easily visualize your DSPy modules
Link to full video in second tweet
2
3
35
@FrDesouche
@tanguy_france
Être raciste c'est vraiment coûteux: entre financer des policiers tueurs et acheter des journaux pour passer un message symbolique, vous avez de l'argent dis donc. De quoi vous plaignez vous donc ?
31
0
31
Wow already 1 customer and a $10 /m for !
This is already better than 0 lol.
I just added a new chapter on advanced modules and retrievers in DSPy ! Go watch
#buildinpublic
#dspy
DSPy to the moon 🔥🔥🔥🚀🚀🚀
This is awesome! I am planning to democratize the use of DSPy even more on my LMS platform
I am currently working on a comprehensive tutorial to make DSPy accessible to a lot of AI engineers / developers.
Create an account
3
4
44
5
4
32
@abhi_agarwal4
@sama
Man you should learn about switching costs and consumer inertia. People tend to stick to a product or service out of habit rather than seeking out new options. For LangChain and Llama-index, they will probably survive by providing the same abstractions across LLM providers
3
2
30
@scienceisstrat1
@Noahpinion
@BrankoMilan
@JoePompliano
@wolfejosh
Life expectancy.
Who is really richer? 😄
5
0
31
@BetterCallMedhi
The only thing you need to get funded in France hahaha. You don't even need to have a history of building stuff.
There are exceptions but the rule is clear.
3
0
30
Dive into my latest article where I showcase how I supercharged my AI Financial Analyst with
#LangSmith
! From initialization to in-depth tracing, discover the future of AI monitoring and debugging. 1/2
#buildinpublic
@LangChainAI
@hwchase17
1
6
27
@africatechie
assez étonnant coe tweet. vs transférez la responsabilité de la réussite du stage sur le stagiaire uniquement. Le stagiaire doit être accompagné, il est là pour apprendre et l'entreprise doit l'y aider. Le stagiaire doit faire des efforts aussi. Mais il doit être bien accompagné
8
1
28
New chapter available on the DSPy full course on : Synthetic Prompt Optimization! Enjoy !
0
8
28
By popular demand, I have now merged the two approaches successfully. You either come with some initial data, or just define a pydantic model. In both cases there is a place for you in the DSPy familly ❤️
@PaulLockettkpb
thanks for the suggestion
@krypticmouse
@ndzfs
Herumb’s is dataset-centric. Franck’s is signature-centric, but could be program-centric even. So that can suggest a natural separation between the two.
1
0
5
2
4
26
@Carnage4Life
This is why so many things are wrong in the corporate World.
I think someone being polite, respectful and just doing their job should be enough. But that's just me.
1
2
27
Hey
#buildinpublic
community, it's finally done, and I am ready to present the "Discute" app—a web app with the ambition to allow you to chat with your knowledge base (starting with your PDFs).
Powered by
@pinecone
, with RAG design inspiration from both
@llama_index
and
6
0
26
Yes indeed, amazing fireside chat with
@lateinteraction
and
@hwchase17
on DSPy, LangChain and prompt optimization.
Join us here:
2
6
27
@BetterCallMedhi
> "En 2022, la société n’a vendu que pour 568 000 euros de produits finis et a enregistré près de 90 millions de pertes."
😂😂😂 Mais qui sont ces founders et qui sont les VCs qui soutiennent une startup avec une telle perf ??? Comment peux tu lever 316m en 2020 puis ne pas
1
4
27