Lev Konstantinovskiy Profile
Lev Konstantinovskiy

@teagermylk

1,766
Followers
3,265
Following
59
Media
764
Statuses

Head of Engineering building AI voice agents

Berlin
Joined December 2007
Don't wanna be here? Send us removal request.
@teagermylk
Lev Konstantinovskiy
9 months
I am organizing a small #covidsafe techno party on Sun, 14 Jan in Berlin😊: Looking for volunteers to help with #PlusLife testing, driving, setting up air filters and the sound system. DM me if interested🙏. #TeamVorsicht
Tweet media one
63
138
588
@teagermylk
Lev Konstantinovskiy
6 years
NLP style transfer is here. Best paper at #emnlp2018
Tweet media one
14
133
536
@teagermylk
Lev Konstantinovskiy
2 years
"Everyone is sick" story now has some data behind it. Dec 2022 England had the most Antibacterial and Penicillins prescribed per capita since 2014. Infectious meds, Amoxicillin on par with the worst month on record. No increase in antifungals. Graphs+code
Tweet media one
5
99
263
@teagermylk
Lev Konstantinovskiy
5 years
Literally all I do as an NLP engineer: No. No. That's not a clearly defined annotation label. Just try labelling one document. No. Transfer learning still needs labelled data. No. Your corpus is too small for this method. No. Just use a regex. No, regex is still AI. No.
@rafalab
Rafael Irizarry
5 years
Literally all I do as a statistician: No. No. That's not the definition of a p-value. No. Trending towards significance is not a thing. No. No pie charts! That only works if data is normal. No. That's logistic regression not AI. No. Your "novel" method was invented in 1918. No.
90
1K
6K
6
61
259
@teagermylk
Lev Konstantinovskiy
9 months
Things I learned from 6 months of organising in-person covid safe events in Berlin.
Tweet media one
14
74
255
@teagermylk
Lev Konstantinovskiy
5 years
Correctly labelled T-shirts from @manwhohasitall while running text labelling training at @pydataLondon with @Agata_Anastazja
Tweet media one
1
20
144
@teagermylk
Lev Konstantinovskiy
9 months
The ticket is €30 to cover the costs and allow me to break even:  -€225 venue -€115 air filter rental and transport -€150 testing For those that can’t afford it I can offer a few reduced/free tickets. Just DM me.
2
5
93
@teagermylk
Lev Konstantinovskiy
9 months
So... we lost the venue for this Sunday’s techno party. 🙁 Anyone know another place in Berlin that can host us on short notice? We bring our own sound system and air filters but we need 1. Outdoor space to test 2. Indoor space to dance 4. Sun or Sat night 5. Fits 50 people
@teagermylk
Lev Konstantinovskiy
9 months
I am organizing a small #covidsafe techno party on Sun, 14 Jan in Berlin😊: Looking for volunteers to help with #PlusLife testing, driving, setting up air filters and the sound system. DM me if interested🙏. #TeamVorsicht
Tweet media one
63
138
588
3
42
81
@teagermylk
Lev Konstantinovskiy
9 months
We have a venue. 🎉 They reached out themselves and it is absolutely free of charge. 😃 This community is great. 🫂 See you Sunday! 17:00 Test 18:00 Dance
@teagermylk
Lev Konstantinovskiy
9 months
So... we lost the venue for this Sunday’s techno party. 🙁 Anyone know another place in Berlin that can host us on short notice? We bring our own sound system and air filters but we need 1. Outdoor space to test 2. Indoor space to dance 4. Sun or Sat night 5. Fits 50 people
3
42
81
6
11
66
@teagermylk
Lev Konstantinovskiy
6 years
Only 2 weekends left to watch @fastdotai videos for the London Data Science Journal Club meetup discussion of #ulmfit transfer learning for NLP. This time with cameo appearance by one of the authors @seb_ruder . Updated study links in the event description:
1
6
52
@teagermylk
Lev Konstantinovskiy
9 months
Looking for volunteers for the covid-safe Techno Party next weekend. Can you help testing, driving, setting up air filters or the sound system?
1
11
49
@teagermylk
Lev Konstantinovskiy
9 months
@ABu7314 @EndemieRebellen Thanks! If that is more your thing, I also host live singer-songwriter performances. Some people travel from other cities and stay for a weekend. This was in November:
5
2
45
@teagermylk
Lev Konstantinovskiy
9 months
The entrance is 30 euros to cover the costs: €225 venue, €115 air filter rental and transport, €150 testing. ​If you can't afford it, you can join for free, just DM me.
0
3
31
@teagermylk
Lev Konstantinovskiy
6 years
Slides for my talk at @pydatalondon this morning. "Using transfer learning for automated factchecking". Thanks to @alex_conneau for open-sourcing very useful #infersent embeddings!
@FullFact
Full Fact
6 years
How is Full Fact using natural language processing to supercharge factchecking? Our NLP engineer @teagermylk explains this Saturday @pydatalondon
Tweet media one
1
7
10
0
10
30
@teagermylk
Lev Konstantinovskiy
9 months
@Nephrodite1 👋 Freut mich zu hören! Wir werden im Juni und jeden Monat eine Party veranstalten. 😀
3
1
27
@teagermylk
Lev Konstantinovskiy
8 years
New document similarity using Word Mover's Distance in #gensim | @pydatalondon talk slides
@saeedamenfx
Saeed
8 years
cool Gensim library for identifying similar sentences/reviews, will look at! - @teagermylk @pydatalondon
Tweet media one
0
2
7
1
15
27
@teagermylk
Lev Konstantinovskiy
9 months
@gedankenknaeul Herzlichen Dank! Ich hoffe wirklich, dass mehr Veranstaltungen dies tun. Ich habe eine kleine Anleitung geschrieben, damit es einfacher ist, die Maßnahmen umzusetzen
3
9
24
@teagermylk
Lev Konstantinovskiy
6 years
Awesome use of @gensim_py and @spacy_io !
@durand101
@dldx.org on bsky
6 years
My @puddingviz piece on women in @UKParliament has made it to the @infobeautyaward shortlist! It's my first ever data viz / ML project so I'm kinda in shock! Read it here: And please vote for it: Cheers! #iibawards
3
9
38
2
6
26
@teagermylk
Lev Konstantinovskiy
7 years
. @mattilyra Coherence is very useful during training. See this WIP by @parul1sethi #pydatabln #gensim #meaning
0
6
23
@teagermylk
Lev Konstantinovskiy
1 year
@karpathy Hmm for me that takes the fun away! I somehow like writing my Anki cards myself, choosing what to highlight. That helps me to remember later. And the formatting is easy with the Obsidan Anki plugin.
3
1
23
@teagermylk
Lev Konstantinovskiy
5 years
Looking forward to our text annotation workshop today 15:00 at #pydatanyc . Thanks to all the volunteer game masters who came to training. Special thanks to Mars Lee for making the logo, and to @w4rnertech for making the website.
Tweet media one
2
1
21
@teagermylk
Lev Konstantinovskiy
7 years
Stoked - joining @FullFact in London as #NLProc engineer. Amazing mission & years of fact-checking expertise. Looking for Data Engineer now.
@FullFact
Full Fact
7 years
Full Fact is building world-first automated factchecking tools. Read about it here:
Tweet media one
4
51
67
2
2
22
@teagermylk
Lev Konstantinovskiy
9 months
@LC_UK_Action Thanks for the supportive words. We will dance for you! I hear it a lot, sadly: "I wish I knew about your party before I got LC" :(
1
0
21
@teagermylk
Lev Konstantinovskiy
5 years
Hi, we are looking for a Python Developer / Data Engineer in London or Bangalore. Would be grateful for recommendations.
Tweet media one
4
18
21
@teagermylk
Lev Konstantinovskiy
8 months
Hi all, if you would like to know more about covid safe events in Berlin please follow the new dedicated @safehourberlin account. Two new events announced in this thread!
@SafeHourBerlin
Safe Hour Berlin
8 months
Our January techno party was great. Thanks to everyone who volunteered and participated. Here are a few photos.
Tweet media one
Tweet media two
Tweet media three
2
2
8
0
7
20
@teagermylk
Lev Konstantinovskiy
2 years
Thanks to @openprescribing for open sourcing their code.
1
1
17
@teagermylk
Lev Konstantinovskiy
6 years
1/6 Here is an explainer thread accompanying the paper
@FullFact
Full Fact
6 years
How can we use machine learning to automatically detect factcheckable claims in live debates? We improve on current claim detection systems by 5% in our paper for #emnlp2018 #NLProc Congratulations to @teagermylk @olly_is_price @meandvan @arkaitz
Tweet media one
2
15
45
1
4
16
@teagermylk
Lev Konstantinovskiy
5 years
Looking for Game Masters to help run a Dungeons & Dragons - style text annotation workshop at #pydatanyc next week. We are already 5 and need 5 more.
Tweet media one
3
5
15
@teagermylk
Lev Konstantinovskiy
6 years
Next London Data Science Journal Club will be on ULMFit and newer transfer learning in #NLProc . We have to do it as even non-NLP people heard of them! Join us on 23 Oct "Easy life in NLP: only 100 examples to train a classifier with Transfer Learning"
1
7
15
@teagermylk
Lev Konstantinovskiy
7 years
#pyconru was fun! Giving a joint talk only 3 hours after meeting @menshikh_iv offline for the first time was surprisingly easy.
Tweet media one
Tweet media two
Tweet media three
0
2
14
@teagermylk
Lev Konstantinovskiy
7 years
I have successfully forked myself! @menshikh_iv is the new maintainer of @gensim_py since June. He just ran an awesome sprint and talk!
Tweet media one
Tweet media two
1
5
14
@teagermylk
Lev Konstantinovskiy
9 months
@LC_UK_Action masked people having fun together is so controversial to the “living in fear” stereotype
2
4
14
@teagermylk
Lev Konstantinovskiy
6 years
1/6 Here is the explainer thread accompanying our #emnlp2018 paper "Towards Automated Factchecking: Developing an Annotation Schema and Benchmark for Consistent Automated Claim Detection"
@FullFact
Full Fact
6 years
How can we use machine learning to automatically detect factcheckable claims in live debates? We improve on current claim detection systems by 5% in our paper for #emnlp2018 #NLProc Congratulations to @teagermylk @olly_is_price @meandvan @arkaitz
Tweet media one
2
15
45
4
5
13
@teagermylk
Lev Konstantinovskiy
8 years
Finally a new #word2vec example! Nirvana - male_vocals +female_vocals =Hole :) #nycmachinelearning
Tweet media one
0
9
12
@teagermylk
Lev Konstantinovskiy
9 months
@Seadop2 Yeah, hope to see you there another time. We do an event every month - techno, live music, board games.
1
0
12
@teagermylk
Lev Konstantinovskiy
6 years
Looking forward to speaking at #emnlp2018 #fever workshop tomorrow! Will talk on how we defined and automated Claim Detection to help our factcheckers at @FullFact
Tweet media one
1
0
12
@teagermylk
Lev Konstantinovskiy
7 years
Looking forward to speaking at @pydatalondon conference in April about applications of #InferSent sentence embedding to factchecking
2
5
12
@teagermylk
Lev Konstantinovskiy
8 years
Kicking off 1st day of #gensim sprint at #pyconindia "Learn Machine Learning by improving Gensim tutorials"
Tweet media one
0
2
11
@teagermylk
Lev Konstantinovskiy
7 years
Honoured to give a #pycon2017 lightning talk on word movers distance
@ehmatthes
Eric Matthes
7 years
Omg now a want to be an astronomer and a mathematical linguist! #PyCon2017
0
0
4
1
3
11
@teagermylk
Lev Konstantinovskiy
7 years
GloVe #wordembedding in Science "Semantics derived from corpora are biased"
0
4
10
@teagermylk
Lev Konstantinovskiy
7 years
Notebook for my "Get the similarity you need" talk at #pydatabln later today
1
2
10
@teagermylk
Lev Konstantinovskiy
9 months
@Datentante @Nephrodite1 Zu diesen Terminen werden wir eine Veranstaltung durchführen. Wir wissen noch nicht genau, was der Inhalt sein wird - Techno, Live-Gitarrenmusik, Brettspiele oder alles andere, was die Leute anbieten wollen! 4. Februar 2024, 17. März 2024 und 7. April 2024
0
1
10
@teagermylk
Lev Konstantinovskiy
5 years
#spacyirl was the best conference ever :)
@seb_ruder
Sebastian Ruder
5 years
@yoavgo on (some of the) missing elements in NLP. Future vision: humans writing rules aided by ML. #spaCyIRL
Tweet media one
Tweet media two
Tweet media three
Tweet media four
6
32
130
0
0
9
@teagermylk
Lev Konstantinovskiy
8 years
Excited to speak at @pydatalondon on Sunday. Doc classification with #gensim . "Word embeddings for fun and profit"
0
3
10
@teagermylk
Lev Konstantinovskiy
9 months
@JustineGSwaab Thanks. I love the stylish Dräger's too. We don't require FFP3 but we offer bitrex fit testing on site. And encourage it as a party game while we wait outside for the test results.
0
0
11
@teagermylk
Lev Konstantinovskiy
6 years
So glad to see someone explain so well how NLP projects actually happen. And glad to know that here @FullFact we follow the best practices in model design and annotation. Thanks for the great talk!
@honnibal
Matthew Honnibal
6 years
In NLP and ML we talk a lot about models and optimization. But this isn't where the battle is really won! I've been trying to explain my thoughts on this lately. Big thanks to @PyDataBerlin for a great event. 📺 Slides:
2
61
191
1
2
10
@teagermylk
Lev Konstantinovskiy
7 years
Every other time slot is an #NLProc talk at #PyDataLdn Thx to the NLP organizer @marcobonzanini for that! Next year call it PyNLPDataLondon
0
3
9
@teagermylk
Lev Konstantinovskiy
8 years
Dear £800 a day Spark contractor, you have been Bangalored!
Tweet media one
0
0
9
@teagermylk
Lev Konstantinovskiy
9 months
@LC_UK_Action Yeah, of course. I took a video at our last one but it was really dark and you could just see blurry white masks and hear the music :) I hope to find a professional videographer and photographer to make a "“Having fun doesn’t kill people. Greed and indifference do” poster.
2
0
9
@teagermylk
Lev Konstantinovskiy
6 years
Pretty cool! Multimodal embedding meets transfer learning for search at an art collection. #infersent sentence embeddings + DeViSE. Code on github.
@hmpim
Harrison Pim
6 years
This is what has been occupying my mind for the last few months - a deep-learning based image search which ACTUALLY LOOKS at the images, rather than matching your query against unreliable captions
3
17
40
0
1
9
@teagermylk
Lev Konstantinovskiy
6 years
Yesterday 40 people chose to spend a sunny Thursday evening indoors, reading about imbalanced learning. It's great to see the growing community of people learning together! Thanks to Ash for teaching me about "Focal Loss" and to @cristohowlo for organising
Tweet media one
1
0
9
@teagermylk
Lev Konstantinovskiy
9 months
@silviafrankeao1 @ABu7314 @EndemieRebellen Großartig. Du bist die Person, die wir brauchen. 😃
0
0
9
@teagermylk
Lev Konstantinovskiy
8 years
Code sprint for #gensim in full swing @bitspilanigoa
Tweet media one
Tweet media two
Tweet media three
0
1
8
@teagermylk
Lev Konstantinovskiy
2 years
@docbrummer I got my initial in person appointment with a therapist in Berlin cancelled after I asked questions on risk mitigation and said I will wear a mask. "I need to see your face for this work" Glad they recognise their limitations, and will still resent them when I see them on zoom.:)
2
0
5
@teagermylk
Lev Konstantinovskiy
9 months
@PJ17862 Unfortunately, no. Hybrid events are the hardest to do well. A good video stream in a dark room especialy. Instead, recommend this fully virtual coviding club
Tweet media one
1
1
8
@teagermylk
Lev Konstantinovskiy
7 years
"The only things that should divide us are significant indentation and tabs vs spaces." Glad to be at #pycon2017 #codeofconduct
Tweet media one
0
2
8
@teagermylk
Lev Konstantinovskiy
10 months
@brownecfm I know several stories of households where infection didn't spread beyond the first case. They do pooled pluslife molecular tests a couple of times a week and have the masks/hepas/space to isolate.
1
2
10
@teagermylk
Lev Konstantinovskiy
5 years
If you are at @pydatalondon conference on Sunday, come along to our role-playing workshop with @agata_anastazja @bhargavvader @lopusz @w4rnertech . It won't be very technical but it will involve dice and annotating text. What could be more fun?
1
7
8
@teagermylk
Lev Konstantinovskiy
9 months
@DelacroixMimi Es tut mir leid, von deinem Postcovid zu hören. Vielleicht sind Brettspiele oder Live-Musik für dich besser geeignet? Wir fahren ein paar Leute mit wenig Spoons zu und von der Veranstaltung.
2
0
8
@teagermylk
Lev Konstantinovskiy
9 months
@_Kun3_0 Sorry to hear. It is too much for one person to organise this. The survival is in the community.
1
0
7
@teagermylk
Lev Konstantinovskiy
5 years
Finally common sense for DS from the people who gave us great software dev processes.
@martinfowler
Martin Fowler
5 years
post: @dtsato , @arifwider , and @intellification start describing the technical components of continuous delivery for machine learning. First up: discoverable data and reproducible model training
1
41
79
1
0
7
@teagermylk
Lev Konstantinovskiy
6 years
Or maybe we need better human performance metrics. SWAG human evaluation is only on 100 samples, so the human accuracy of @rown (the expert) annotating his own dataset is in the range of 71-99% when sampling from a binomial see @GaelVaroquaux 's
@seb_ruder
Sebastian Ruder
6 years
It's amazing how fast #NLProc is moving these days. We have now reached super-human performance on SWAG, a commonsense task that will only be introduced at @emnlp2018 in November! We need even more challenging tasks! BERT: SWAG:
Tweet media one
Tweet media two
8
87
294
0
3
7
@teagermylk
Lev Konstantinovskiy
1 year
@brownecfm And sadly no Novavax offered.
0
0
7
@teagermylk
Lev Konstantinovskiy
2 years
Come work with us.
@_inesmontani
Ines Montani 〰️
2 years
If you work with JavaScript and are interested in ML tooling, come and argue with web browsers for us! ⚔️✨ Explosion is hiring a Senior Full-Stack / Front-End Developer to work on our upcoming product, Prodigy Teams. Details & application:
Tweet media one
1
19
67
0
0
6
@teagermylk
Lev Konstantinovskiy
9 months
@SadRobot1980 Thanks. We did some experiments with frozen samples and for the purpose of finding the positive person in a pool of 3 unrelated people, that works. There are graphs in the doc.
1
0
7
@teagermylk
Lev Konstantinovskiy
9 months
@HootMouna You are very welcome! We do one event every month.
0
0
6
@teagermylk
Lev Konstantinovskiy
5 years
@dirk_hovy You can have the next of both worlds
0
1
6
@teagermylk
Lev Konstantinovskiy
6 years
5/6 Best performing model was #infersent sentence embeddings by @alex_conneau
Tweet media one
0
1
6
@teagermylk
Lev Konstantinovskiy
8 years
@deliprao Thanks for bringing WordRank to our attention with your blog post!
@gensim_py
Gensim
8 years
New Gensim wrapper for WordRank embedding. “Crowned” is most similar to “king”, not #word2vec ’s “Canute”.
Tweet media one
1
33
63
0
1
6
@teagermylk
Lev Konstantinovskiy
3 years
February Online London Data Science journal club will be on Text-based NP Enrichment, plus SpanBERT
1
1
6
@teagermylk
Lev Konstantinovskiy
9 months
1
0
6
@teagermylk
Lev Konstantinovskiy
6 years
4/6 Annotations were crowdsourced with the help of @FullFact 's awesome volunteers using Prodigy by @explosion_ai
Tweet media one
0
1
6
@teagermylk
Lev Konstantinovskiy
3 years
@tallinzen Reminds of some branches of string theory called by critics "not even wrong" as they can explain any kinds physical phenomena. Is it even science when it is hard to know if the model is wrong or not? With rules it is clear what their limits are.
0
0
6
@teagermylk
Lev Konstantinovskiy
7 years
Очень рад делать доклад в воскресенье на @PyConRu с @menshikh_iv "Gensim тематическое моделирование для людей"
Tweet media one
0
2
6
@teagermylk
Lev Konstantinovskiy
1 year
@MsMacrophage Naive question - is there evidence of Antibody-dependent enhancement, like in this thread?
@arijitchakrav
Arijit Chakravarty
1 year
The preprint (which we expect will be published fairly soon) outlines some additional unpleasant consequences of the current “let it rip” strategy for Covid. The TL;DR version is that there is a plausible risk of a dengue-like situation eventually developing for covid (2/)
2
24
128
1
0
4
@teagermylk
Lev Konstantinovskiy
7 years
Understanding cannabis slang with #gensim #word2vec trained on 70 mln tweets. By @RTI_Intl
0
2
5
@teagermylk
Lev Konstantinovskiy
6 years
@cboutaud Thanks! Greatly enjoyed your keynote too. By the way, this is the paper about sarcasm detection with eye tracking
Tweet media one
1
0
5
@teagermylk
Lev Konstantinovskiy
7 years
@gensim_py
Gensim
7 years
We'll sprint again like we did last summer! Looking fwd to sprinting Mon-Tue at #PyCon2017 in Portland,USA
0
1
5
1
3
5
@teagermylk
Lev Konstantinovskiy
3 years
My favourite part of #nlpproc by far. And in the past I could sense clients (and even reviewers) getting frustrated with this "unexpected" stage existing at all and taking time.
@yoavgo
(((ل()(ل() 'yoav))))👾
3 years
*a lot* of the time went into defining, re-defining, and refining the task itself, to make it amenable to annotation with high agreement and high coverage, even among ourselves. while keeping it useful. we did many iterations there. many.
1
1
8
0
0
5
@teagermylk
Lev Konstantinovskiy
7 years
@manwhohasitall The world just has to progress to understand our hormones. I don't really mean to be aggressive in the morning and cry in the evening.
0
0
5
@teagermylk
Lev Konstantinovskiy
9 months
@LC_UK_Action The poster to echo this historical one from Act Up
Tweet media one
2
0
5
@teagermylk
Lev Konstantinovskiy
6 years
Hoping for convergence of academic and journalistic terms soon... A great tensor embedding paper by @vagelispapalex named "fake news detection" could be "style differences of established news vs the ones with mis/dis-information" as it doesn't focus on factchecking. #KDD2018
1
0
5
@teagermylk
Lev Konstantinovskiy
9 months
@jual1977 Thanks! Some people drive from other cities and stay overnight with local folks in a bedroom with a hepa. And a "pluslife before breakfast" of course :)
0
0
5
@teagermylk
Lev Konstantinovskiy
9 months
@pipingbob @_Kun3_0 Jetzt in Tempelhof
@teagermylk
Lev Konstantinovskiy
9 months
We have a venue. 🎉 They reached out themselves and it is absolutely free of charge. 😃 This community is great. 🫂 See you Sunday! 17:00 Test 18:00 Dance
6
11
66
0
0
3
@teagermylk
Lev Konstantinovskiy
7 years
100 data scientists walk into a bar: "Can we fit?" Barman says: "Absolutely not. If you could, you'd have predicted and transformed."
2
3
5
@teagermylk
Lev Konstantinovskiy
3 years
@burkov I never just use off the shelf embeddings. That's unprofessional! I fine tune on 100 domain docs that I have ranked myself after a half an hour call with the client.
1
0
5
@teagermylk
Lev Konstantinovskiy
9 months
@strugglingjulie Hi. There are several people from NRW interested - maybe you can drive together? Also, our next event is board games on 4 Feb - you are very welcome to come for that.
0
0
5
@teagermylk
Lev Konstantinovskiy
1 year
@brownecfm Please accept my condolences.
0
0
1
@teagermylk
Lev Konstantinovskiy
9 years
Thx @ds_ldn @futurecitiescat best organised hackathon ever. Air expert help on tap, hosted ipython, fast wifi, even pizza-free food! #AQHack
0
1
4